Skip to main content

Fast Company Technology

What we learned from 250,000 public workers about AI adoption

3 weeks 1 day ago

As generative AI sweeps across industries, some workers using it are reporting increases in productivity. But inside government, where those benefits could translate into better public services, many employees are still unsure how the technology applies to their jobs, how to use it, or whether they are allowed to.

That’s one takeaway from our research at InnovateUS, a nonprofit learning initiative that has trained more than 250,000 government workers in AI, digital technology, and innovative skills. They work in all 50 states and the District of Columbia, in jobs ranging from public safety and benefits administration to legal services and executive leadership. Our aim is to give people working in state and local government accessible opportunities to develop skills that help them serve the public, and ultimately deliver services worthy of residents’ trust. 

During no-cost skill-building workshops across the country and on the web, public servants trade experiences about what works and what doesn’t. Participants have explored using artificial intelligence to improve benefits delivery with New York City’s health commissioner and learned from current and former state-government leaders how to introduce workers to using and even building AI. Recently, one state leader in Colorado described how he is using generative AI to improve performance reviews; he reported better feedback, greater worker satisfaction, reduced turnover, and time savings. These exchanges help public professionals learn from one another’s experiments and consider what they could adapt to their own agencies.

Learning together online also gives us an opportunity to listen systematically to people’s experiences with AI. That listening underpins the first report from our Observatory of Public Sector AI, drawing on 106,300 precourse survey responses and 67,002 postcourse responses from participants in our introductory generative AI course, collected between June 2024 and May 2026. The findings reveal a substantial gap between the headlines around generative AI and everyday working life in government.

What keeps public workers from using AI

Among public-sector respondents entering our foundational generative AI course, 51% had never tried the technology. Another 14% had experimented with it personally but weren’t yet using it at work. While technology companies race to build autonomous agents, many workers are still trying to understand what AI has to do with their jobs, why they are being asked to use it, and how to use it well. They’re also not sure if they’re allowed to use it at all.

The findings complicate the stereotype of government employees as resistant to change. Among respondents who weren’t using generative AI, 28% said they didn’t think it applied to their work, 23% didn’t know how to use it, and 19% didn’t know whether they were allowed to use it at work.

Managers face similar uncertainty. Among managers who weren’t using AI, 24% didn’t know how, and 22% didn’t know whether it was permitted. Employees look to supervisors for guidance, but those supervisors may need help themselves.

Listening to the workforce gives leaders something concrete to address. Providing access to a tool or publishing a policy leaves substantial work undone. Employees need relevant examples, opportunities to practice, and managers prepared to support responsible use.

An overview of findings from the 2026 Observatory of Public Sector AI [Image: InnovateUS]

To turn AI use into public value, we need to help employees connect new skills to specific service problems. This might look like reducing wait times when people call about health or unemployment insurance, using AI to make it cheaper and easier to engage the public in policymaking, or even knowing how to regulate and buy AI to reduce dependence on any one vendor. And we need to give them the support to change how work gets done. Systematic workforce training creates an opportunity to build that capability to listen, learn and deliver, and to understand whether it makes a difference. 

Beginners report the largest gains

One of our most encouraging findings concerns people starting with the least familiarity. In the matched responses, learners who entered with no reported AI knowledge showed the largest gains in reported knowledge. Those who began with no confidence also experienced particularly substantial improvements in confidence.

This parallels research in the private sector. A study of private sector customer-support workers by academics at Stanford and MIT found that less experienced and lower-skilled workers gained the most from an AI assistant. That study measured workplace performance; ours measures reported learning. Together, these studies suggest that organizations should pay attention to people who are just getting started. Concentrating support on enthusiastic early adopters risks overlooking workers who could benefit considerably from an introduction.

Findings from before and after trainings in Washington, DC. [Image: InnovateUS]

We also found a striking relationship between confidence and the possibilities people could envision for generative AI tools. After training, highly confident respondents planned to apply AI to an average of 3.8 tasks, compared with 1.5 among those with low confidence. That association does not establish that confidence causes broader use. It suggests that learning involves more than acquiring facts about a technology. People also need to feel capable of identifying and trying appropriate applications.

Confidence alone, however, is an incomplete measure of success. Someone can feel more capable without becoming more accurate. Training must develop the judgment to question an answer, verify a source, protect sensitive information, and recognize when AI is inappropriate. We need to assess those capabilities alongside confidence.

From better tasks to better services

When asked how they hoped to use AI, participants readily described activities such as drafting reports, creating presentations, and summarizing documents. Fewer responses connected those activities to outcomes such as making services more accessible or improving public participation.

Our questions partly explain that pattern. We asked about specific tasks, and brief open-ended responses offered limited insight into what people hoped to accomplish through them. We have begun collecting voice-based feedback and  are adding questions that probe whether learning is helping employees improve services.

Private-sector research underscores why the distinction between individual learning and organizational outcomes matters. In McKinsey’s 2026 global survey, 80% of respondents said AI improved their own productivity, yet only 37% reported a positive impact on their organization’s earnings. Helping individuals work faster does not automatically improve an organization’s performance.

For democratic institutions, the corresponding test is whether time saved translates into better public services, more engagement with residents, and more effective problem solving. Does faster document processing reduce the wait for benefits? Does clearer communication help more people complete an application? Does using AI to analyze public feedback lead to a better decision? Ultimately, do people trust government more? Training needs to help employees make those connections, and evaluation needs to tell us whether they hold up in practice.

What we still need to learn

Our first report captures an early stage in a much longer process. Its surveys are voluntary, and participants are a large but not representative sample of the government workforce. The results measure reported knowledge, confidence, and intended use immediately before and after a course. They do not establish whether people retain what they learn, apply it consistently, make fewer mistakes, or deliver better services. Without a comparison group, we also cannot isolate the effect of training from other influences. The scale of the dataset helps us examine patterns across professions and organizations, though it does not remove those limitations. 

Our opportunities to learn more are accelerating. As our learner population grows and we expand the range of people teaching and learning from one another, we can investigate which approaches help different workers and what support they need after an introductory course.

Places such as Washington, D.C., which has adopted a whole-of-workforce training approach, offer an especially valuable opportunity. Studying learning across an organization can help us understand how individual skills interact with managerial support, access to tools, and agency rules.

The most exciting next step is connecting evidence about learning with administrative measures of service performance and other data on trust in government. With appropriate privacy protections, we could examine whether training is associated with shorter wait times, more accurate case resolution, fewer application errors, easier access to services, and more outreach to residents 

Those studies will require careful design. A shorter queue could reflect staffing changes, new software, or lower demand. Following teams over time and comparing their experiences could help us understand whether, how, and under what conditions learning contributes to more effective, efficient and engaged institutions. 

We have encouraging evidence that public professionals can gain knowledge and confidence around AI, including those starting with little familiarity. Now we need to follow what happens next as we work with governments to roll out new training that teaches people how to use AI in their specific role and to work in new ways. 

For government leaders, that means treating workforce learning as an investment whose results deserve sustained attention. This would include things like offering practical training, preparing managers to support it, giving employees opportunities to apply what they learn, and measuring what changes for the people they serve.

At a time when many people are skeptical of AI and uncertain about how it will benefit them, government has an opportunity to demonstrate what the technology can accomplish in the public interest. AI should not be a technology whose benefits accrue primarily to the companies building it or the people who can afford it. 

By equipping public servants with the skills to use AI responsibly and creatively, we can put powerful new capabilities in the hands of people whose job is to serve others. The measure of success will not be how many government employees use AI or how much time they save, but whether people experience better services, have more opportunities to participate in decisions, and see government solving problems that matter to them. Making those kinds of differences in people’s lives could make AI worthy of their trust and, more importantly, help rebuild their trust in government.

The Grace Hopper Celebration was more than a conference. Its cancellation should be a wake-up call

3 weeks 1 day ago

This week, thousands of women and nonbinary technologists were supposed to walk into the Grace Hopper Celebration in Anaheim, California. Instead, the doors are closed. The tech industry named some of its most powerful AI chips after Grace Hopper. This year, it could not sustain the conference that bears her name.

AnitaB.org canceled this year’s celebration after individual registrations and sponsorship commitments fell short of what it needed to run the event. The celebration was first held in 1994, co-founded by computing pioneer Anita Borg. Last year, it drew nearly 15,000 attendees from more than 60 countries. The loss is bigger than a few days on the technology industry’s calendar. The Grace Hopper Celebration was not merely a conference or a symbolic celebration of women’s contributions. It was part of the industry’s career infrastructure, and when budgets tighten, convenings are often the first to go.

Confidence and information

For students and early-career technologists, that infrastructure matters. Research from Girls Who Code and the Burning Glass Institute, set to be released next month, shows the entry-level door for software engineers has narrowed. From early 2022 to early 2026, the share of software-developer postings open to people with two years of experience or less fell from 19.3% to 14.1%, a 27% decline. This makes a conversation with a recruiter or mentor, or having a community of peers, more important than ever. Those people can give someone the confidence and information to remain in a field where they are underrepresented.

At Girls Who Code, we see and hear stories like this often among our 425,000 alumni. Technical skills are essential, but skills alone don’t create careers. Young people also need someone to open doors, make introductions, and help them imagine where they belong.

Those doors are becoming harder to find as artificial intelligence reshapes the technology workforce. PwC’s “2026 Global AI Jobs Barometer,” which analyzed 2.4 million U.S. entry-level job postings, found that the roles most exposed to AI are now seven times more likely to require traditionally senior skills like leadership and judgment. We are asking young people to arrive with experience no one has given them the chance to build.

More barriers than doors

Women made up 37% of computer scientists in 1995; today, it’s about a quarter. The next generation is ready, though. When Girls Who Code surveyed 2,328 teens for our “Agency in the Age of AI” study, we found that girls aren’t disengaged from AI. They’re discerning about it, and they use it more when they feel ownership over how they use it. The same study found that by around age 17, girls are losing confidence that they can make a difference in AI. It’s easy to see why, when they see more barriers than doors.

The Grace Hopper Celebration is one piece of a broader ecosystem that supports young women’s professional development in tech. Its loss this year signals a need to reexamine how people enter the technology workforce. The technology industry has an opportunity to rethink how it supports the next generation of talent by reducing barriers to entry, creating more accessible pathways into emerging fields like AI and cybersecurity, and expanding access to career opportunities. By investing in young talent and strengthening connections among employers, educators, and communities, we can help ensure more young people can build meaningful careers in technology. These actions can create more durable, accessible pathways into technology that remain open year-round and have a wide reach.

This is a talent strategy. The people most likely to identify a technology’s blind spots are often those whose experiences have been missing from the rooms where it is designed. When researchers at the University of Washington tested AI résumé screeners in 2024, the tools consistently favored names associated with men.

Creating a platform

AI is transforming nearly every industry, and companies are searching for people who can use technical tools while exercising judgment, creativity, and an understanding of how technology affects others. They cannot afford to recruit from an increasingly narrow pool.

That’s why, together with our founder, Reshma Saujani, Girls Who Code will share presentations from speakers who were scheduled to take the stage at the Grace Hopper Celebration across our social media channels throughout the week. Their insights, from engineering to career advice, were meant for thousands of attendees. They should still reach those attendees, and many more who could never have made the trip.

The women and nonbinary technologists who’d planned to attend in Anaheim shouldn’t have to wait another year for those opportunities. The conference is not happening this year. We cannot let the on-ramps into technology narrow along with it.

This Texas startup is turning oil well waste into EV batteries

3 weeks 1 day ago

Across the Permian Basin, the vast, mineral-rich swath of land in West Texas that supplies much of the nation’s crude oil, derricks pump out massive amounts of oil, roughly 6 million barrels a day. 

But that amount pales in comparison with the amount of dirty, brackish water that’s also extracted during oil drilling. Drilling in the Permian Basin also unearths roughly 26 million barrels of water a day, about the same daily water use as all of New York City. This water is often black-gray from oil residue, and has a funky odor from underground bacteria. 

Traditionally considered a by-product of energy production, this water, concentrated seawater from prehistoric oceans, is mineral rich. The Houston-based startup Element3 found that this so-called wastewater is a rich source of lithium, and endeavored to filter it to create a domestic source of this high-demand mineral. (Lithium is the third element on the periodic table, hence the company name.) It’s a creative concept that, in a poetic twist, turns a waste product of petroleum drilling into the raw materials for EV batteries and a clean energy future. 

“It’s a wonderful circular economy, waste-to-value story,” said the company CEO and founder, Hood Whitson.

Founded in 2021, Element3 is commercializing U.S. Department of Energy research from the Oak Ridge National Laboratory on ways to filter this oil by-product, and is working across Texas to figure out ways to scale up production. The firm has raised a little over $20 million and has 15 employees, per PitchBook. For years, DOE researchers have been looking into what Whitson called “unconventional brine mining,” finding ways to extract lithium from California’s inland Salton Sea, for example, or pull rare earth metals out of industrial waste sites. 

Aged for hundreds of millions of years, the Permian Basin water is five times as salty as today’s oceans, boasting lithium as well as magnesium and strontium. Whitson discovered the properties of this water, and realized that energy companies had built up a massive infrastructure to move it; in effect, paying for the infrastructure for extraction and returning filtered water to the larger system. It seemed like an ideal way to meet the rising demand for lithium, and build a domestic source for the metal, which is mostly processed in China. The U.S. imported more than $16 billon worth of lithium-ion batteries from China in 2024.

“This is one of our nation’s largest untapped natural resources,” said Whitson. “It’s really a grand opportunity.”

Currently, the wastewater gets diverted through a massive series of pipes to be used by the energy industry for fracking. Element3 sets up a processing plant adjacent to these pipes that diverts, filters, and returns the wastewater to the existing pipe system, eliminating the need to dispose of or move it. 

Element3 has a 6,000-square-foot commercial extraction facility in Midland near the oil fields. To filter the water, the facility uses a proprietary medium, a specialized direct lithium exchange unit that uses absorptive materials to soak up the lithium ions. The result is lithium chloride, which the facility processes via carbonization and electrolysis into lithium carbonate powder and lithium hydroxide, two key raw materials for batteries. The site, which opened in February, is the first new lithium extraction facility in the nation since the 1960s. It can produce up to 3,000 tons a year. Compare that with the only existing U.S. source of lithium, a mine in Silver Springs, Nevada, which produces about 5,000 tons annually.

“This is a new opportunity to grow another source of good-paying energy jobs, and proof that our nation’s energy companies can continue to evolve with energy demand,” said Whitson. 

If you set aside the fossil fuel extraction that is already happening—a big thing to set aside—just the Element3 process of extraction from that wastewater is a lower-emissions means of generating lithium than the existing U.S. mine, as well as many foreign processing facilities in places like Chile, Whitson said. Other companies see riches in reusing this water: Iofina just opened a West Texas facility to create iodine from this by-product.  

Whitson sees companies like his as part of a larger shift happening across Texas. The energy industry, while traditionally dominated by oil and gas, has long embraced an anything-and-everything approach, making big strides in wind, solar, and geothermal. Often, especially in the case of geothermal, it uses a skilled oil and gas industry workforce to grow. 

“We strongly believe in the electrification of everything,” said Whitson. “We’re now able to layer a critical materials industry on top of what West Texas has already built.”

When to buy Apple’s accessories (and which ones to buy cheaper elsewhere)

3 weeks 2 days ago

Dropping $1,200 on a brand-new iPhone only to have an Apple Store salesperson casually hand you a $49 silicone case and a $39 charging brick feels like a minor mugging.

Yes, Apple’s accessory margins are legendary. The company has built a multibillion-dollar side business selling $19 polishing cloths and $39 plastic dongles, banking on the assumption that you will blindly buy official gear for everything just to keep your new device working safely.

But that assumption is wrong.

At the same time, going bottom-barrel cheap for certain power components can result in slow charging speeds, excessive heat, or degraded battery health. Knowing where to spend your money and where to save comes down to a few clear rules.

Here’s a practical breakdown of which Apple accessories justify the official markup, and which ones you should buy third-party for a fraction of the price.

Buy official MagSafe wireless chargers and power banks

There are a few specific categories where buying official Apple gear is genuinely worth the price tag, mostly where precision engineering, thermal safety, and input latency come into play.

MagSafe wireless chargers are a prime example. The official $39 Apple MagSafe charger contains proprietary hardware modules that communicate directly with iOS thermal management. It delivers fast wireless charging up to 25 watts while throttling output if your phone gets warm.

Most $15 magnetic knockoffs on Amazon are merely basic 7.5-watt Qi chargers with a magnetic ring glued inside. They generate excessive heat, charge your phone at half speed, and degrade long-term battery health.

Apple Pencil and iPad Magic Keyboard

The same rule applies to the Apple Pencil and iPad Magic Keyboard. An official Apple Pencil offers sub-millisecond latency, tilt sensitivity, and pressure tracking that generic $20 active styluses simply cannot replicate.

For the $300 (and up) Magic Keyboard, Apple’s Smart Connector passes power directly through the hinge without requiring separate Bluetooth pairing or charging dongles, and the glass trackpad remains unmatched in daily feel compared with plastic third-party alternatives.

Buy elsewhere USB-C charging cables

When it comes to physical protection, textiles, and basic copper wiring, however, skip the Apple Store and head straight to a third party.

Standard USB-C charging cables are a classic case of Apple overcharging for a fragile product. Apple’s white cables are notorious for yellowing, fraying at the collar, and splitting after a year of travel.

Trusty third-party brands like Anker, Ugreen, and Spigen sell double-braided nylon USB-C cables rated for 100-watt power delivery and 30,000 bends for less than half the price of an Apple cable.

Cases and screen protectors

Protective cases and screen protectors are another area where third-party manufacturers consistently win. Apple’s official $49 silicone cases famously peel at the corners after a few months in a pocket, and their clear cases are notoriously slippery.

Third-party case makers offer air-cushioned corner protection, grippy textures, and built-in MagSafe rings on the cheap. Furthermore, paying Apple $40 in-store for a Belkin glass screen protector is painful when a $10 three-pack on Amazon includes an auto-alignment tray that guarantees a bubble-free installation in seconds.

Watch bands

Apple’s $50 woven nylon and silicone watch bands carry a criminal markup. Third-party woven nylon bands match the stretch, comfort, and lug fit of official bands so closely that you can buy a four-pack of different colors for a fraction of the cost of one official strap.

AirTag keychains and mounts

As for AirTags, paying $35 for a leather holder for a $29 tracking puck makes no financial sense. Third-party retailers are flooded with rugged, waterproof, screw-lock AirTag mounts that protect the tracker far better than Apple’s open-face leather loop—and most cost around $10.

The golden rule for outfitting your Apple gear is straightforward: Buy official for internal power management and precision inputs, but buy third-party for physical protection, textiles, and charging cables. Sticking to vetted tier-one brands like Anker, Spigen, and Ugreen allows you to fully kit out your new setup while keeping a substantial chunk of change in your bank account.

Town CEO Jean-Denis Greze on why most people still don’t know what to ask AI

3 weeks 2 days ago

AI agents are powerful, but most ordinary people don’t yet use them. Town wants to change that. Led by CEO Jean-Denis Greze, playfully referred to as Town’s mayor, the company announced a $108 million Series B on October 19, co-led by Forerunner Ventures and Menlo Ventures, at a $1 billion valuation.  

Greze cofounded the San Francisco startup with Tony Vincent in late 2024 after serving as CTO of Plaid, where he built up the engineering team from about 20 people to 350.

Town’s betting it can make AI more accessible and practical for small-business owners and mid-market firms still looking for day-to-day AI benefits. 

Fast Company spoke with Greze about why most people still treat AI like a fancy search engine, where AI assistants can be most helpful, and why he’s working harder than he ever has before. The conversation has been edited for length and clarity.

ChatGPT and Claude can already connect to other apps. Why do people need stand-alone AI assistants like yours?

Could Claude or ChatGPT do most of what Town does? I think the answer is, most of it, yeah, they can. But most people who use ChatGPT or Claude aren’t using those tools to do what people in tech describe as agentic work, which just means using them to actually do things for you as opposed to researching things for you. They use it more as a Google Search replacement.

I think the biggest reason is that most people don’t really know what to put in the chat box. Putting someone in front of an empty box and asking them to write what to do, especially when the capabilities are changing all the time—that’s not onboarding.

No one’s figured out how you take a mainstream person and get them to use these products to their full potential. That’s the market opportunity. And we may be wrong, by the way. Maybe distribution just wins, and Google, Microsoft, and OpenAI end up winning. As a startup, that’s the risk you’re taking on.

What’s a use case that surprised you?

There’s a small travel firm at a vacation destination where people rent cars and gear, like a bike, for the day. The business is totally run over email. If something’s damaged, the employee takes pictures and sends an email to the central processing person.

Now there’s a little workflow that at the end of the day takes all the emails with the pictures, looks at the CRM for who the people are, and drafts the emails. They were drafting 10 to 12 of these emails every day, and it wasted 25 minutes; now it takes them 5 minutes. No one’s building SaaS for that small business. 

What should we call this category of tool? An AI assistant? A chief of staff?

We call our product a Townie. If you use the word assistant or the phrase chief of staff, you have a human mental model for what an assistant does. I think what this technology does today and what it’ll do in five years will probably defy these expectations. And no one knows what a chief of staff does outside of CEOs of Fortune 500 companies.

I actually think the right analogy for what we’re building is more like a browser or a device. When the App Store came out, suddenly a percentage of what you used to do in your web browser, you started to just do in your phone. Adding AI assistance is the same. 

The percentage of the time you interact with information in a browser or your phone will just go down now forever, and the percentage that you do by talking to an AI will go up. The future, I think, is voice for sure.

What’s next for Town?

I’m very excited about live voice. You could imagine in the morning you’re commuting to work and you have in your ear a world-class executive assistant that’s telling you, “Hey, you got seven emails overnight. There’s two of them that are urgent. On this one, if you tell me this, I can help draft the reply. The other one is going to require you to move a meeting later today. Do you want me to come up with some options?” It’s just talking to you, and you can interrupt it.

I’m also excited about what we call agent to agent. You could say, “Who at the company is really close with someone doing R&D at this other company?” Your agent will talk to other agents at your own company, and it’ll learn that Bob plays tennis with someone there. But it won’t tell you that. It’ll just be like, “Hey, Bob appears to be closely connected to that person. Do you want me to ask them if they can do the intro?” Leveraging the network of everyone within a company in a way that’s privacy-preserving is huge.

Should people connect their financial accounts to an AI agent?

If you’re a consumer, it is very likely that you connect everything about you, your medicals, your financials, to the agents. It’s just too beneficial. It’ll tell you that your credit card rate’s too high, that you should apply for a different one. It’s free money.

But the privacy concern is real. I don’t think you want to use an assistant where you’re not convinced it’s making decisions that are in your best interest. These products only work in a world where you think they truly have your best interest.

Can you trust those services?

If you’re paying for it, you can. If you’re not paying for a service, you’re not the customer. You’re being monetized in some other way.

We made the choice very early: We like consumer use cases, and they’re subsidized by business use cases. We charge businesses, and it’s not a cheap product; it’s not $5 a month. Our main plan is $50 a month.

What’s the hardest part of running Town?

The world’s too fast. I’ve never seen anything like it. It’s nerve-racking. Software had high gross margins because there was a really high fixed cost to copying it, and there was a real advantage to being early in market. By the time someone could copy you, you’d have built the next feature.

With AI, competition moves so fast and good ideas are copied so quickly. Things that would take nine months to copy are copied in a month or two.

So the pressure you feel to execute and to work a lot—I’m 47, but I’ve not worked this hard [before]. Psychologically, it’s tough, because you’re either ecstatic because you shipped something before OpenAI did, or you’re in fear that there’s a tweet from Sam Altman that puts you on your back foot. So that’s super stressful.

How do you use Town outside of work?

I’ve told Town to research food blogs and newspapers that write about food in San Francisco. I have a date night with my wife every Sunday night. So every Monday night, it looks at the next month and gives me four tables that I can make a reservation at that are classics, and then finds four that are new that have been mentioned in food blogs. I like food. I just don’t have the time [to do the research myself].

Anthropic was beaten to the punch by a salad dressing company

3 weeks 2 days ago

What do you do when you think your company’s technology may fundamentally alter the fabric of the economy—or rend it apart entirely?

Anthropic, which is widely rumored to be approaching a record-setting IPO, apparently has an answer: Cut the public in on a piece of whatever comes next.

Last week, the company announced that it would dedicate 5% of its shares to a company-owned public benefit trust.

If Anthropic’s products are as lucrative as its sky-high valuation suggests, those shares could be worth billions on day one—and could grow to be worth hundreds of billions over time.

By my calculations, if Google had done the same thing before its 2004 IPO, the shares it set aside for the good of humanity would be worth $184 billion today.

So Anthropic’s plan is a bold one. Depending on where you sit, you’ll likely see it either as a genuine effort to give back to the world, or an attempt to shake regulators off the company’s back—or stop pitchfork-wielding mobs from storming its lovely new office building in SoMA if things get really bad.

And though the scale is unprecedented, it’s not a new idea. In fact, Anthropic was beaten to the punch by a movie star turned salad dressing tycoon. 

Yet Anthropic’s approach has far more concerning implications.

Exploiting Yourself

By age 57, actor Paul Newman was doing pretty well for himself. He had starred in almost 50 movies—including classics like Cool Hand Luke and Butch Cassidy and the Sundance Kid—and was on the path to an eventual Oscar.

Somehow, he’d also managed to find the time to win second place in the grueling 24 Hours of Le Mans auto race, among other wild and wonderful accomplishments.

When someone suggested he slap his iconic blue-eyed visage on bottles of salad dressing to boost sales, Newman apparently thought, Are you crazy?

He said the idea felt like “exploitation.” But then Newman realized that Hollywood had always been exploiting him for his good looks and boyish, all-American charm. Why not do it for a good cause?

And why not do it as fully as possible? In Newman’s words, “Why not, really, go to the fullest length, and the silliest length, in exploiting yourself, and turn the proceeds back to the community?”

So that’s what he did. At nearly 60 years old, he launched Newman’s Own—a salad dressing company that would donate 100% of its proceeds to charity. The company eventually expanded, putting his name and likeness on tons of other products in the same act of “exploitation.”

Today, as any Whole Foods shopper can attest, Newman’s Own still exists (and makes some pretty respectable salsa). 

More importantly, the idea worked. The company has reportedly generated over $600 million in charitable donations since Newman founded it in 1982—a massive sum for a company the actor started “on a bit of a lark.”

Altruism, Full Circle

At the time, Newman seems to have been mostly messing around. It’s unlikely that he knew his company would prove so successful in its charitable mission.

But a far more concrete version of the philosophy Newman intuited now drives the actions of companies like Anthropic.

Newman called it “shameless exploitation in pursuit of the common good.” Today, we call it effective altruism. As The New York Times explored earlier this year, it’s a mindset that’s absolutely pervasive in the world of frontier AI.

The idea is essentially to amass as much wealth and power as possible, with the goal of giving it all away, or otherwise turning your influence to the good of humanity.

It sounds nice on paper. But the philosophy, as practiced nowadays, can have a dark side, as when it allegedly drove “crypto king” Sam Bankman-Fried to defraud his way to riches, the better to give them to causes of his choosing.

It could easily have a dark side in the AI world, too—with potentially far graver consequences.

Even if Anthropic gave away all its shares à la Newman, the drive to make those shares as valuable as possible for humanity could motivate the company to scrimp on safety or to callously build things that, by its own admission, risk destroying the world.

Anthropic’s desire to share some of its massive wealth is probably a good one. But as the humans stuck holding the bag if AI goes horribly wrong, we shouldn’t let Anthropic’s gesture blind us to the much bigger questions of AI safety, creative rights, and societal impacts that still surround its work.

It’s a good thing for the powerful and influential to consider their impact on the world. But Anthropic should take a page from Newman’s book and remember that altruism isn’t just about money, nor do the positive ends always justify the means.

There’s not much point in setting aside cash to help humans, for example, if your technology has inadvertently stamped them out.

The AI infrastructure race: I’ve seen this movie before

3 weeks 3 days ago

Back in 2022, when the auto industry was fixated on autonomous vehicles, I was cofounder and chief technology officer at Luminar, where our lidar technology was helping enable the next generation of self-driving cars. Speaking at an automotive conference, I remarked how exciting it was to have a front-row seat to one of the most significant technological transformations in the industry’s history.

Afterward, an automotive executive pulled me aside.

“Jason,” he said, “you don’t have a front-row seat. You’re part of the show. What you’re doing for autonomy, and how you’re forcing us to think, is exactly what’s needed.”

That conversation stayed with me ever since, because it captured something larger than autonomous driving. When an industry is being transformed by technology, the companies that built the old infrastructure are not always the ones that define the new one. In autonomy, some of the companies that pushed furthest were technology companies like Tesla, Waymo, and Zoox; not the automakers that had dominated transportation for a century.

I believe we are watching the same pattern emerge again.

Today, I’ve shifted my focus from optics for autonomous vehicles to optical networking for AI infrastructure. And once again, technology companies are confronting infrastructure that was built for a different era. Rather than wait for traditional power and telecommunications providers to catch up, they are taking the problem into their own hands.

That is the shift that matters. The companies building artificial intelligence are no longer just customers of infrastructure. They are beginning to become infrastructure companies themselves.

Just as they did during the autonomy race, the technology companies aren’t waiting for the traditional infrastructure industries to solve the problem.

The benefits of urgency

When autonomous vehicles were supposed to be the next big thing, the tech innovators and the auto companies attacked the problem from different directions. Tech companies threw themselves into lidar (light detection and ranging), radar, computer vision, and AI software with genuine urgency.

Many of the traditional automakers approached it differently. They invested enormous sums—tens of billions of dollars collectively—through internal programs, acquisitions, and partnerships. Ford created Argo AI and later Latitude AI. General Motors built Cruise. Volkswagen invested heavily in Argo AI and Cariad. Toyota, Mercedes-Benz, and Nissan each launched major autonomous initiatives of their own. Despite these investments, many of the highest-profile programs were scaled back, reorganized, or shut down as the complexity of bringing true autonomous driving to market became clear.

Meanwhile, companies like Waymo and Zoox, born in the technology industry rather than the automotive industry, continued advancing because they reimagined the entire system around software, AI, and compute.

The result is visible on the roads today: Waymos and Zooxes that don’t come from Detroit.

Tech race 2.0

Having helped build one technological transformation from the inside, I’ve learned to recognize the early signs of the next one. I’m watching a similar game unfold again, though with different players and far higher stakes.

The driving force this time is AI. Hyperscalers, the Googles, Microsofts, Amazons and Metas of the world, are racing to build the data centers that will power the AI economy. Power has emerged as one of the defining constraints on AI infrastructure. These facilities are extraordinarily power hungry, and the grid isn’t keeping up.

Just as they did during the autonomy race, the technology companies aren’t waiting for the traditional infrastructure industries to solve the problem.

They’re investing directly in power generation, private optical networks, advanced cooling, and the technologies needed to connect geographically distributed AI campuses into what increasingly functions as a single AI factory.

Microsoft signed a 20-year power agreement that is helping make possible Constellation’s restart of Three Mile Island. Elon Musk has pursued natural gas generation for xAI’s data center campuses for much the same reason. When a single AI campus can justify billions of dollars in dedicated energy investment, the economics of infrastructure begin to change, and the companies building AI begin to look a lot like infrastructure companies themselves. Texas has even moved toward requiring many new large data centers to bring their own generation capacity, a recognition that the grid alone cannot keep pace with AI’s growth.

None of this is because utilities are failing. They are doing exactly what society has asked them to do: provide reliable, affordable electricity within a highly regulated framework. Permitting, environmental review, transmission planning, and rate-case proceedings all take years. AI infrastructure is being deployed in months. The mismatch isn’t competence. It’s speed. While utilities work within those constraints, the hyperscalers are increasingly building around them.

Telecommunications is experiencing the same shift. Much of today’s carrier infrastructure was designed for an internet built around consumers, enterprises, and mobile devices. AI changes the traffic patterns entirely. Rather than wait for carriers to evolve, hyperscalers are building private optical networks between their own campuses and purchasing networking equipment directly from manufacturers. Once again, they’re becoming builders of infrastructure instead of merely customers of it.

A new generation of innovators is asking a fundamentally different question: What if the underlying architecture itself needs to change?

History rarely repeats itself exactly, but it often rhymes.

We can’t afford to play catch-up

This isn’t an argument for utilities or telecommunications providers to step aside. Quite the opposite. Their role will become even more important. But institutions built around long investment cycles, regulated returns, and legacy infrastructure cannot be expected to move at the pace of AI. The companies creating the demand are increasingly becoming the companies creating the infrastructure.

I’ve had the unusual privilege of helping build technology through two major infrastructure transitions. What I’ve learned is that technology almost always advances faster than the institutions surrounding it. Eventually those institutions adapt. But rarely before innovators have forced them to rethink what’s possible.

The show has already started. The question isn’t whether that infrastructure will change. It’s who will build it.

Booking holiday flights? Here’s how to know when to pounce.

3 weeks 3 days ago

As anyone who travels frequently knows, airfare costs around the Thanksgiving and Christmas holidays can be significantly higher than those for the same route at other times of the year. It’s little surprise why: the year-end holiday period sees millions traveling to visit family and friends, creating demand for flights; the more demand, the more airlines can charge.

The good news, according to travel experts at Google Flights, is that if you are planning to fly for Thanksgiving or Christmas, you still have a small window of time to get the lowest airfare possible for the winter holiday 2026 period before prices go up—but you have to act fast.

When is the best time to purchase airfare for Thanksgiving 2026?

If you know you are going to be traveling by plane this Thanksgiving—and you haven’t purchased your tickets yet—you shouldn’t wait much longer. As a matter of fact, the ideal time to purchase your Thanksgiving flights may have just passed.

According to a blog post from the travel experts at Google Flights, the search giant’s flight aggregator and comparison tool, there is a period before the holiday when flights are generally at their lowest. This period is generally 21 days to 57 days before departure.

So, if your Thanksgiving flight departure date is actually on Thanksgiving—November 26 this year—that means airfares will be at their lowest between September 30 and November 5 (adjust those dates accordingly to your actual Thanksgiving departure plans).

The prices during this period, of course, are not fixed. They’ll change by the day based on demand and other metrics. On some days, the same flight could be more expensive or cheaper. 

But Google Flights notes that there is one day on which its data shows that Thanksgiving flights are generally at their lowest. That day, for 2026, is 34 days before departure.

Using flying on Thanksgiving morning as an example, the 34-day prior mark means airfare was most likely lowest on Friday, October 23. If you’re planning on flying on the Wednesday or Tuesday before Thanksgiving, the cheapest fares for you were likely on October 22 or October 21.

Does that mean you’re out of luck if you haven’t already purchased your Thanksgiving flights? Not necessarily. Remember, Google says flights should remain at their lowest until about 21 days before departure this year. That gives you until the first few days of November to purchase your airfare—but the sooner, the better.

When is the best time to purchase airfare for Christmas 2026?

If you haven’t purchased your Christmas airfare yet, you should do it soon. That’s because, according to Google, the lowest prices for Christmas flights occur much farther out than they do for Thanksgiving flights.

According to Google, the lowest prices for Christmas flights are 33 to 67 days before departure. Given Christmas is on December 25, and assuming you would be flying that morning, that means that the lowest prices for Christmas flights would generally be found between October 19 and November 22. Of course, since most people travel before Christmas Day, those dates shift even earlier.

If you’re looking for the day with the statistically lowest Christmas airfare prices this year, Google says the “lowest prices hit 56 days before departure” in 2026. If you are traveling on Christmas Day, that means the day you are likely to find the cheapest flights will be this coming Friday, October 30.

Can popular airfare hacks save me even more money on Thanksgiving and Christmas flights?

When it comes to airfare, the internet is full of advice that purportedly shows you how to save even more—from using VPNs to buying tickets on a Tuesday to clearing your browser’s cookies.

But could these so-called hacks actually save you money on your Thanksgiving or Christmas flights?

It’s unlikely, no matter what time of year you are flying. That’s according to Google Flight’s James Byers, who is a group product manager at Google Search. Last year, I asked Byers if Google Flights data showed whether any of the three most common bits of conventional wisdom about buying airfare held up—at least when searching for flights on Google’s aggregator.

Byers revealed that clearing cookies or using Incognito mode didn’t impact airfares on Google Flights. When it comes to using VPNs to try to trick airlines into thinking you’re in a country where prices may be lower, Byers said it may work—until you have to turn over your billing address for your payment method. At that point the VPN-adjusted local fare will likely no longer be available to you. 

As for buying flights on a Tuesday, supposedly the day of the week airfare is cheapest, Byers said there was some truth to that. But airfares were generally only about 1.3% lower than on the most expensive day, Sunday.

As for this holiday season, Google Flights data show that the best time to buy, if you want the lowest airfare possible for your Thanksgiving or Christmas trip, is quickly approaching its end.

How to turn off Google’s AI Overviews—and reclaim search sanity

3 weeks 4 days ago

We’ve got two competing realities when it comes to tracking down info online these days:

  1. Getting answers has never been easier, between all the artificial intelligence chatbots in front of us and the unavoidable smearing of “AI Overview” answers at the top of regular Google search results.
  2. Getting accurate information has never been more challenging—thanks to the fact that all that AI-generated gobbledygook is flat-out wrong a shockingly high amount of the time (anywhere from 10% on the lower end to 50% and up, when it comes to more niche and specialized info).

To use some highly technical terminology, that ain’t great.

BUT: There’s an easy, powerful workaround to skip all those accuracy-challenged internet search “shortcuts” and go back to the basic, traditional web search of yesteryear—without any AI and without any costs or compromises.

Today’s Cool Tool is one you’ll absolutely want to hang onto.

This tip originally appeared in the free Cool Tools newsletter from The Intelligence. Get the next issue in your inbox and get ready to discover all sorts of awesome tech treasures!

Your AI-free Google search workaround

Few folks are aware—or remember—but back when Google first started gunking up its search results with all sorts of AI drivel, it built in a detour to avoid all of that and get pure, simple, clean web results without any of the added noise.

➜ That workaround is still available, if you know how to activate it. And it does more than just strip out the AI Overview additions. It also removes shopping boxes, info boxes, related news suggestions, and pretty much all the other distractions that litter search results these days.

⌚ It’s incredibly easy to do, too—we’re talkin’ two seconds of active effort.

And the difference really is staggering. I mean, look at this side-by-side comparison showing standard Google search results for a random query compared to search results with all the extra noise removed:

The standard 2026 Google Search view, at left, and the stripped-down and AI-free alternative, at right.

Quite the contrast, no?

✅ And, again, there really isn’t much to this, once you know where to look:

  • The simplest way to get pure and simple Google search results without AI Overviews or any other additions is to perform your searches through a site called UDM14​. All it does is send your query directly to Google with the right code at the end to remove all the extra junk and show you only standard web results—without any added tracking or other chicanery.
UDM14 makes a minimalist Google Search experience especially easy.

Using the UDM14 site is by far the simplest way to do this, but hey, you’ve got options.

And whichever way you go, you’ll find yourself basking in the joy of AI-free, clutter-free search results for the first time in ages.

Can I get a collective “hallelujah”?!

  • UDM14 is ​a regular ol’ website​ that works in any browser, on any device, and doesn’t require any downloads or installations.
  • It’s free to use; there’s a single small and unobtrusive ad on its main landing page.
  • The site doesn’t track searches, require sign-ins, or collect any manner of personal data.

Treat yourself to all sorts of brain-boosting goodies like this with the free Cool Tools newsletter—starting with an instant introduction to an incredible audio app that’ll tune up your days in truly delightful ways.

What AI leaves out of the answer still counts

3 weeks 4 days ago

There’s a contradiction emerging in how publishers get paid when AI uses their content. The licensing deals that OpenAI, Meta, Amazon, and others have been signing these past few years appear to be slowing down, at least for the publishers still waiting on one.

Publishers are speaking about this openly. Digiday hosted a London town hall recently, and one publishing executive said discussions with AI companies have slowed down. Also, Katya Gorchinskaya, a Ukrainian journalist and media consultant who built J·Index, a database of thousands of AI implementations in newsrooms, separately told the Reuters Institute for the Study of Journalism that the market for deals with model makers is effectively closed.

At the same time, AI demand for content continues to go up. I recently spoke with Jonathan Roberts, chief innovation officer at People Inc., for The Media Copilot podcast, and he revealed that the number of different bots that try to scrape content from the company’s media properties (including People, InStyle, and Investopedia) now numbers in the tens of thousands every day. The company blocks all but 38 of them. At that Digiday London event, Fiona Spooner, managing director of consumer revenue at the Financial Times, said one trial account tried to access about a million articles in eight hours from 700,000 IP addresses.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/media-copilot.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/fe289316-bc4f-44ef-96bf-148b3d8578c1_1440x1440.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to The Media Copilot\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for The Media Copilot. To learn more visit \u003Ca href=\u0022https:\/\/mediacopilot.substack.com\/\u0022\u003Emediacopilot.substack.com\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/mediacopilot.substack.com\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91453847,"imageMobileId":91453848,"shareable":false,"slug":"","wpCssClasses":""}}

I also see this in my GEO client work: the media properties often start from the position that having bots hammer your site all day is cost-prohibitive, so they block them outright. It makes sense, but if the end goal is to monetize your content in AI systems, the slowdown in licensing agreements suggests an all-or-nothing stance won’t give you much leverage.

All of this points to an evolving approach to payment: metered pricing. The industry has already been moving in this direction, but the question that doesn’t yet have a clear answer is what those meters should measure.

Paying for what gets used

Metered pricing has been around for a while. TollBit was one of the first to offer a “pay per crawl” model, directing bots to a gate where they could elect to pay a price set by the site, or go away (the company says its technology is now installed on more than 7,000 publisher sites). Cloudflare followed suit last year with a fanfare announcement that it would block AI bots by default for new sites on its network and provide the tools for publishers to charge by the crawl.

Things have evolved since then. There’s now more energy behind a “pay per use” model, meaning a content provider would only get paid when their content contributes to an answer, not at the moment it’s scraped. Cloudflare itself made that switch in July, with Ceramic.ai and You.com as its first partners, and Google and Microsoft are testing similar models. Cloudflare’s program is still in beta, but Lara Cohen, Cloudflare’s VP of strategic partnerships for media, creators, and AI, told me the company is working toward a system that alerts publishers when a buyer wants their content and lets them accept, decline, or negotiate.

The logic is informed by data: According to research from AirOps covering more than 548,000 pages, only about 15% of what ChatGPT retrieves ends up in its final answers. This is intuitive and matches human experience: When you do research to inform a report, you don’t typically use all of it.

However, we should be careful of stretching the stat too far: Just because chatbots cite only 15% of what they look at, that doesn’t mean the 85% left out is worthless. In fact, uncited sources still shape answers. An Ahrefs study of 1.4 million prompts found ChatGPT leans heavily on Reddit to understand topics while rarely giving it credit. That material might also be exactly what’s needed for a different query, so there is value in having it available.

This is why, when determining value (and thus payment), who’s paying is as important as how much. If it’s an AI app answering a user’s question, it only wants to pay for what’s used. If it’s whoever builds the index that app draws on, there’s value in having the widest corpus of content available.

There’s actually a third party here: the middleman. And the question is even more relevant for them. As I’ve explored before, a whole industry of AI data brokers has emerged over the past couple of years, with hugely valuable companies like Exa, Parallel, and Tavily leading the pack. These companies crawl the internet, then sell access to the corpus to buyers—which are often AI developers. Parallel, valued at $2 billion after a $100 million round in April, says more than 100,000 developers use its tools, at companies like Harvey, Notion, and Opendoor.

Roberts doesn’t think much of the business model. “These businesses take content from the internet for free and sell it for money,” he told me. “They do not create things, they resell things.”

Pricing varies, but they bill customers by the request, which amounts to charging per delivery. Exa charges $1 for every 1,000 pages of full text, and Parallel’s Extract is $1 per 1,000 URLs. In publisher terms, that’s about a tenth of a cent per article.

Some of the services don’t pay publishers at all, but that’s starting to change. Since May, Parallel has partnered with several publishers, including The Atlantic, Fortune, and PR Newswire, on payments through a program called Index. The payouts depend on the measured value of the publication’s contribution to any work Parallel’s agent does on behalf of customers. That incentivizes unique, hard-to-replace content, since it’ll earn more.

The case for an index fee

Although the market for middlemen clearly needs more structure, Parallel’s approach is a real step toward the sophisticated payment system the industry needs. That said, it’s one part of an economy with several layers. Assuming the system can scale, paying publishers only when their content is used is a tough sell since it gives them nothing up front, and it doesn’t address what I mentioned earlier: the value of having the content in a larger corpus in the first place.

For example, say you had a query about a company’s layoffs, and the agent looks at three sources: Source A, an original report with the layoff numbers, which nobody else has; Source B, a wire story that repeats A’s numbers; and Source C, background on the company’s finances. The agent would likely use A, discard B, and maybe lightly incorporate bits of C. Money would flow accordingly.

But B is still valuable to validate Source A, and it’s important to have in cases where there is no Source A. And the next time, Source B might be the one with the original report. To incentivize the widest number of sources, a nominal, flat index fee (not the same as pay-per-crawl) could go a long way toward convincing publishers to participate in the broader system.

That’s less hypothetical than it sounds. Cohen told me that on Cloudflare’s platform, each buyer defines “use” for itself: “Some define this as when content is indexed, while others define it as when content is used to generate answers.” In other words, some buyers are already paying for the library.

As we’ve seen with the bigger licensing deals, this kind of system might be difficult to scale, but that can be addressed by scaling the fee to how much demand there is for a publisher’s content, not how big the publisher is. That’s important because smaller publishers can punch above their weight for the right query. Cohen said the biggest lesson from pay per crawl’s first year was “the tremendous value of long-tail, smaller publishers’ content,” especially local news. “If I’m asking ChatGPT what’s happening today in Duluth, Minnesota, there’s probably only one or two local news sources to pull that information from.” A small paper with no competition is Source A for its whole town.

TIME’s Mark Howard has seen evidence of the library’s value, too. He told me the models appear to store what they scrape, since some TIME articles get more AI referrals than citations, something that shouldn’t happen. And most of the agent crawls TIME sees are of archive stories more than a year old. The library is getting used whether or not anyone pays for it.

Roberts pointed to Spotify as a good analogy for what the industry currently lacks: “We’ve basically got 18 Napsters of content, but there’s no Spotify.” Spotify is actually a great model for this system: It pays rights holders based on their share of streams, but its deals with the major labels have included large advances and guaranteed minimums. A leaked 2011 contract with Sony, for example, showed up to $42.5 million in advances over three years.

In the case of the media, the middlemen would pay publishers a nominal amount for inclusion, then the real money comes in as their content provides actual value for customers.

Google holds the meter

It’s a solid model, but it doesn’t necessarily translate from the “middlemen” market to the mass market. OpenAI already has many licensing deals in place and doesn’t seem interested in other approaches. Perplexity experimented with rev-share models (an ad revenue share in 2024, then Comet Plus, which sends 80% of subscription revenue to participating publishers), but they haven’t had much effect. And the biggest fish, Google, is unique: It’s recently introduced a pilot called News AI for publishers to collect payments for use in Gemini, AI Overviews, and AI Mode, but the pricing is a total black box, and since it has all the leverage, there’s no reason to think that will change.

Which brings it back to the meters. The industry is finally building them, and that’s real progress. But a meter that only counts answers will always undervalue the library, and the library is what makes the answers any good. If publishers are going to keep feeding these systems, someone has to pay for the pantry as well as the ingredients that end up on the plate. Roberts has a simple test for whether the price is right: “If we would create new content for this market, then the pricing is right. If we would not create new content for this market, then the pricing is wrong.” By that measure, we’re not there yet.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/media-copilot.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/fe289316-bc4f-44ef-96bf-148b3d8578c1_1440x1440.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to The Media Copilot\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for The Media Copilot. To learn more visit \u003Ca href=\u0022https:\/\/mediacopilot.substack.com\/\u0022\u003Emediacopilot.substack.com\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/mediacopilot.substack.com\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91453847,"imageMobileId":91453848,"shareable":false,"slug":"","wpCssClasses":""}}

The U.S. Treasury wants expanded restrictions on high-tech investments to these ‘countries of concern’

3 weeks 4 days ago

The U.S. Treasury Department proposed a rule Thursday to crack down on American high-tech investments such as supercomputing and artificial intelligence in China, Russia, North Korea and other “countries of concern.”

The U.S. government has been working for years to deny China and other adversaries investments that could help them boost their military power.

That program began in 2023 when then-President Joe Biden signed an executive order putting restrictions on outbound U.S. investments in sectors such as semiconductors, quantum computing and artificial intelligence going into China. Treasury later established the Outbound Investment Security Program.

In December 2025, Congress passed a law that codified and expanded the program to target more technologies and more foreign countries, including Cuba, Iran, North Korea and Russia.

The inclusion of Russia in the proposed rule has come days after President Donald Trump waived sanctions on Russia following a diesel deal with President Vladimir Putin.

The rule announced Thursday would expand the restrictions to cover investments in hypersonic systems capable of traveling at five times the speed of sound, supercomputing, biotechnology and artificial intelligence, though Treasury used Trump’s favored term for AI, “superintelligence” or SI.

The proposed rule “sends an important signal to markets that Washington is tracking where U.S. capital is flowing and what capabilities it is helping build,” said Craig Singleton, a senior fellow at the think tank Foundation for Defense of Democracies.

The screening of outbound investments “recognizes that capital can accelerate military-relevant capabilities just as surely as technology transfer can,” Singleton said.

Jason Chipman, a former Justice Department lawyer who specializes in cybersecurity, investment rules and export controls as a partner at the law firm WilmerHale, said the new rule “will require more substantial compliance programs to manage risk.”

Treasury’s rule to implement the 2025 law — the Comprehensive Outbound Investment National Security Act — would only apply to investments that exceed $50,000 and would exclude many financial investments, including stock purchases.

Treasury will hear public comments on the proposal before issuing a final rule.

“It’s just the first step in a fairly long process to get to changes which presumably will be more enforcement-minded, to put more teeth behind the outbound investment program,” said former U.S. trade official Ryan Majerus, now a partner at King & Spalding law firm.

In Beijing, the Chinese Foreign Ministry said the U.S. should abide by the principles of a market economy and stop interfering in normal economic and trade cooperation between businesses.

Asked about Washington’s proposal to add biotechnology to the list, ministry spokesperson Mao Ning said that “Overstretching the concept of national security, taking discriminatory measures against specific countries, and artificially restricting trade, economic and investment activities serve no one’s interests.”

She added that China will closely monitor relevant developments and take necessary measures to safeguard its interests.

Since Trump’s first term, the United States and China — the world’s two biggest economies — have been locked in an economic and tech rivalry. They’ve restricted the export of sensitive technology and critical products — such as China’s rare earth minerals and America’s advanced computer chips — to each other and slapped tariffs on each other’s products. Trump and Chinese President Xi Jinping agreed last year to ease tensions while they engaged in talks, a temporary truce they last month extended into January.

But the sparring continues. In July, the Treasury Department issued a $200,000 fine against a U.S.-based company for failing to report to the department a $92,500 investment in a Chinese artificial intelligence and robotics company by one of its subsidiaries.

—Paul Wiseman and Didi Tang, Associated Press

Palmer Luckey and Ben Lamm want to make superdogs for the military

3 weeks 4 days ago

Theatergoers have been fans of the line “Cry ‘Havoc!’, and let slip the dogs of war” since 1599. Now venture capitalists are coming around to the phrase.

A new startup called Warhound has emerged from stealth with $200 million in funding to genetically engineer dogs for the U.S. military and equip them with high-tech gear. Its founders call it a modernization of America’s military working-dog program.

Those founders know more than a bit about both genetic modification and warfare. Warhound was conceived by Colossal co-founder Ben Lamm and Palmer Luckey, co-founder of defense technology company Anduril Industries and founder of Oculus VR.

Colossal is best known for its ongoing attempts to “de-extinct” several animal species, including the woolly mammoth and the dodo. Anduril has received more than $25 billion in government and defense contracts.

“Every day, we ask dogs to put their lives on the line to protect ours. They detect explosives, protect our troops, and run toward dangers that most people are trained to run away from. Yet we’re still relying on outdated approaches to breeding and training these extraordinary animals,” Lamm said in a statement.

Warhound is already working on its first batch of genetically modified dogs, collecting blood samples from what it says are top-performing service animals in military, police, and first responder operations. The company plans to clone those dogs and edit their genes to improve their performance.

The first litter of genetically modified puppies, expected in 2027, will consist primarily of Belgian Malinois, Labradors, and shepherds.

The dogs will also be equipped with “Anduril’s advanced cybernetic Augments, software, and battlefield situational awareness,” according to the company’s announcement. That will include body armor as well as eye and ear protection.

“Man’s best friend deserves man’s best tech,” Luckey said.

The effort also aligns with the Trump administration’s emphasis on domestic production. In announcing the company, the founders noted that 80% of U.S. military dogs, of which there are currently more than 2,200, come from Europe, “creating a critical interdependency on foreign nations for an asset the Pentagon itself calls irreplaceable.”

Luckey and Lamm first conceived Warhound several years ago after a conversation at a moonshot summit. More than half of the dogs considered for military service fail to complete their training, according to the company, driving up costs and making the program less reliable.

Reasons for failure can range from poor eyesight to health problems such as cancer or hip dysplasia. Through gene editing, Warhound hopes to address those issues while enhancing traits such as speed and sense of smell, which are critical to detecting bombs and other hazards. The goal, its founders say, is not to create a new breed of dog but to improve the performance of existing breeds.

Genetically modifying dogs could raise concerns among animal welfare advocates, but Warhound says animal welfare will be a core part of its operations. The company is working with the American Humane Society, which will certify Warhound’s training programs and support efforts to reunite retired military working dogs with their former handlers.

“Working dogs are extraordinary animals whose courage and loyalty save lives,” said Robin Ganzert, president and CEO of American Humane Society. “Success means helping these dogs thrive throughout their lives—from training and active service to the loving, dignified retirement every canine hero deserves.”

David Spirk, a Marine Corps veteran and former senior counselor at Palantir, will serve as Warhound’s CEO. The company will be based in Dallas. Colossal holds equity in the business, while Luckey is a partner but does not have an ownership stake.

AI music’s proliferation on streaming services has been a boon for this Spotify competitor

3 weeks 4 days ago

Like every other music streaming service, Qobuz has been enduring a sudden and violent surge of AI-generated audio, but Dan Mackta, Managing Director at Qobuz, refuses to use the term “AI music.” 

“I’ll call it music-esque content,” Mackta says of the tens of thousands of AI-created songs that hit music platforms daily. “Music goes straight to people’s emotions. It’s not something to fool around with.”

Deezer, another prominent streamer, recently shared that it’s receiving 90,000 AI-generated songs per day. Mackta doesn’t care to find out the exact number coming to Qobuz, but he knows it’s a lot.

“Who asked for this? It’s all just fraud,” Mackta says. “It’s a shame that we have to put so much energy and resources into combating it. But of course we do. We have to. It’s an industry-wide problem.”

And yet, from a certain perspective, “music-esque content” might actually be a blessing in disguise for Qobuz. Since AI slop has been flooding platforms, Qobuz has seen massive growth. The company saw a 45.7% revenue increase in 2025 compared to 8.8% for the overall streaming market with their monthly active users reaching 1.2 million . 

These kinds of spikes can involve numerous factors such as increased brand awareness, product endorsements, and, in 2026, viral internet moments. Qobuz has benefited from all three of late. Its viral moment came when Spotify was running ICE ads, leading to a land rush towards Qobuz. 

Human by design

But something Qobuz has kept at its center as it has grown is its focus on authentic creativity. The French streamer, which expanded into the US in 2019, has clear tenets for approaching the concept: high-quality audio, human curation, original editorial. So, when listeners want to be sure they won’t be subjected to music-esque content, Qobuz is a natural choice. 

As soon as they open the app, they see the words “New Releases Handpicked by Qobuz” front and center. Instead of flaunting an algorithm that tracks their every click, Qobuz shows that its team is working to find the best music possible.

“Music is totally unique, as an art form,” Mackta says. “It really doesn’t work like any other medium, even other arts, in the way that it is a vehicle for human-to-human communication.”

Part of maintaining that vehicle involves regularly cleaning it up. The company has removed some four million AI-generated tracks from its internal search tool because they had no listening activity. It has also excluded about 60% of streams on AI tracks from royalty payouts due to fraud activity. Additionally, in September, the company rolled out a tag that identifies tracks as AI-generated. 

Those are just the latest efforts to keep quality at the center of its offering. When Qobuz started as a digital music store in 2008, it was the first to offer CD-quality lossless (16-bit FLAC) downloads. The company expanded into streaming in 2014, offering lossless playback at launch—a full 11 years ahead of Spotify’s global lossless audio launch. And even now, Qobuz’s lossless streaming is slightly higher quality than Spotify’s. 

Human curation is a weekly process. Qobuz’s global team discusses their favorite music they’ve heard recently, whether it’s an album they got from a label or something a friend recommended. From there, they select which releases are best for the app as a whole and individual regions, such as the US, UK, Australia, and Japan (it operates in 26 countries total). The new releases slider on the app homepage, Album of the Week, and even mood playlists like Summer Reggae are all handpicked.

Finally, Qobuz has an in-app magazine that also pairs articles with artist profiles. Users can read and listen simultaneously, fully immersing themselves in the music. 

Qobuz accomplishes all this with 120 full-time employees. Spotify has over 7,000. 

“The biggest companies that have ever existed are our competitors,” says Mackta. “It would be nice to be able to introduce new features in our apps faster, but we don’t. Instead, we try to be the best online music store with a built-in fanzine that we can be.”

It’s doing some of that work with the help of AI, in line with its AI charter that outlines exactly how AI is used in its business operations. For example, on the recent introduction of the new UI, Qobuz engineers used AI tools to make their back-end work more efficient, but no AI was used in graphic design or anything else creative.

“We can’t be saying one thing about music and another thing about photography. It’s not consistent,” Mackta says. 

The industry is all about it  

Qobuz’s values are resonating with artists, too. Yasiin Bey, the rapper formerly known as Mos Def, who once said the streaming revenue model is “gross,” made his 2009 opus The Ecstatic exclusively available on Qobuz.

“It’s not so crazy to think that starting from a position of respecting the art and the artist actually resonates with people, but it’s as simple as that,” Mackta explains. 

That’s why Emma Wilkes, a professional music journalist, chose Qobuz as the first streaming service she’s paid for. Prior to signing up, she used Spotify’s free tier to discover music but mostly bought her music on iTunes. She especially appreciates Qobuz’s sound quality because she writes a lot in punk, metal, and hardcore.

“When you try and listen to Napalm Death on Spotify, you see how bad Spotify sound quality is,” Wilkes says, going on to mention how Qobuz provides relief from AI as well.

“I don’t think Spotify is doing enough about AI. That’s genuinely really damaging to music, ” she says. “I very much appreciate that Qobuz values human curation as much as it does. It’s about creating and supporting an ecosystem.”

The reality of AI on streaming

Many people may have switched to Qobuz, but Tatiana Cirisano, Vice President of Music Strategy at MIDiA, doesn’t foresee a mass exodus that will drastically affect Spotify’s dominance of the market. 

“The vast majority of people will choose convenience regardless of their personal attitudes or values,” Cirisano says. “People might say, ‘I don’t want to listen to AI-generated music,’ but they may end up listening to it anyway and not notice or care.”

As such, Cirisano relates the rise in Qobuz usership to music superfans such as Wilkes. Spotify has 761 million monthly active users. Only a fraction of that number are willing to buy merch or attend a three-day festival because Spotify, by nature, attracts a less-invested fan. $13 a month provides ad-free access to almost every song in history, and an algorithm constantly feeds them new things.

Pair that with an explosion of AI content, and you have artists competing with AI for a dwindling piece of Spotify’s royalties.  

“Not only were a lot of these AI-generated tracks made using tools that were trained on human catalogs without permission, but that music is now competing for the same listeners,” Cirisano says. “The pot of money that can be paid out to each song and each artist isn’t growing as fast as the number of mouths to feed.”

So, in the wake of AI, when Cirisano consults with streaming services on how to handle this discrepancy, her advice would likely sound familiar to Mackta.

“If we were in a place where people listened to artists and not just songs, where there was context about the artist presented alongside the song,” she says, “I don’t know that AI-generated music would be disrupting the market in the way that it is.”

Change.org wants to turn your TikTok rant into local political action

3 weeks 4 days ago

Most Americans—more than 80%—don’t think that their elected officials care what they have to say. Change.org has been rebuilding its own platform to try to change that.

The platform, founded in 2007, is best known for campaigns that have gone viral nationally, like a petition to get justice for George Floyd in 2020 that got nearly 20 million signatures. Of course, most petitions don’t have that kind of audience. But you don’t need millions of people to make change happen, and over the last two years, Change.org has rebuilt its tools to connect people with decision-makers in their own community.

Steve Dziedzic [Photo: Change.Org]

“The core problem we’re trying to solve is that people just feel unheard,” says Steve Dziedzic, who was named the organization’s new president today. “They don’t think their voice matters anymore in our democracy.”

The site now has “an explicit local focus,” says Dziedzic, who has been leading the platform’s redesign, previously as chief product officer. “That’s a big shift for the organization.” There are practical reasons for this: People care most about local issues, and local petitions are also the most likely to succeed. Even if someone wants to work on a national problem like gun control, he says, it makes the most sense to start with a local law and then expand.

When someone comes to the platform now, it helps shape the message to have the best chance of persuading a local decision-maker to take action. An AI “petition writer” tool on the site starts by asking someone for their story. It helps combine that with a logical argument for the mayor or whoever else you’re asking for change. (Dziedzic, who previously founded a couples’ therapy app that was acquired by Talkspace, says that writing a good petition has similarities with couples’ therapy; it helps to be able to dispassionately talk about what you need.) The company is investing around $100 million to rebuild its platform and add AI.

If someone wants their city to build a bike lane, for example, and they’re focusing on safety and sustainability, the tool could potentially also suggest economic arguments to help convince a mayor that taking away parking spaces won’t harm local businesses.

If someone already posted a rant on TikTok, they can paste a link to that video and the tool will use it to automatically draft a petition. TikTok or Instagram aren’t good places to reach local officials, Dziedzic says, because a mayor can’t tell how many of the people liking a post are actually constituents; Change.org now uses addresses to verify constituents, so city officials get an important signal of when they need to be paying attention. The company also uses a tool to scan social media for complaints that could turn into successful petitions, and the team reaches out to invite new users to participate.

After someone creates a petition, the platform also helps find people who are likely to want to sign. It suggests related subreddits, Facebook groups, and other places to find people who’ve rallied behind the issue in the past. Dziedzic says the feature helps increase the number of signatures by 20%. “Unless you have at least 100 signatures, it’s just really not going to matter to the mayor, unfortunately—they need to know that there’s a nontrivial amount of constituents who care about this,” says Dziedzic. Everyone who signs can upload their own video. The site also helps petitioners pitch local media, since that’s another important way to get local officials to care.

Critically, it also looks up the right decision-maker to target, whether you’re trying to make changes in a park or block a data center. Then it tries to broker a meeting. If a city council meeting is the right venue, it lets you know when the next meeting is happening. You can use the platform to reach out to everyone who’s signed and invite them to attend the meeting too. “It’s starting digitally, but then it ends up in the physical world, helping shape real community,” says Dziedzic.

In the past, some critics have called out petitions as a form of slacktivism; some studies have suggested that when someone signs a petition, they’re less likely to take other, potentially more meaningful actions later. But Change.org’s new structure is geared to nudging people to do more.

The tools are helping users get results. In one recent campaign that started as a TikTok video, a Maryland resident created a petition asking the county for a moratorium on new data center development until residents could get more data about the impacts. More than 10,000 people signed, and the moratorium is now in place. (So far this year, the platform has hosted more than 1,800 petitions about data centers, and more than 3 million people have signed.)

Others have recently used the platform to remove Flock cameras, make intersections safer, and fight for relief from a $3,332 electric bill, among thousands of other things.

The company wants to help society collectively make wise decisions, Dziedzic says, and that could eventually mean going beyond petitions alone—for example, helping decision-makers hear as much as possible from people on both sides of an issue, rather than only the point of view of someone who launches a petition. For now, though, the focus is more straightforward.

“People want to be able to shape their communities, where their loved ones are, where they do life,” says Dziedzic. “By helping people shape their communities in connection with their elected officials, we think that is the shot—that’s my biggest hypothesis for people believing their voice matters again. And it’s most practical, because the local petitions are also the most likely to win.”

The simple joy of a social network most people forgot about years ago

3 weeks 4 days ago

Hello again and welcome back to Fast Company’s Plugged In.

If you owned a smartphone a decade and an half ago and were into cool new apps, there’s a more-than-decent chance you tried one called Foursquare. The best known of a then-new breed of location-based social networks, it let you check into places you visited in the physical world, and see where your friends were checking in.

One Foursquare fan quoted in an October 2009 New York Times article by Jenna Wortham called the ability to find nearby pals for spur-of-the-moment meetups “planned serendipity.” The app also included gamified elements that made it feel less like a utility than a form of entertainment, including badges and the chance to become “Mayor” of your favorite haunts.

For a time, Foursquare had plenty of competition from other check-in apps, such as Gowalla, DeHood, Brightkite, and Loopt (which was cofounded by a guy named Sam Altman). Inevitably, Facebook pounced on the trend, introducing a feature called Facebook Places. However, checking in turned out to be a fad. One by one, most of the contenders faded away—except for Foursquare.

But in 2014, Foursquare the company did something that struck me as an unforced error, and still does: It rebranded its classic check-in experience as Swarm, robbing it of the brand equity it had built up. The Foursquare name lived on through a separate city guide app (discontinued in 2024) and the company’s monetization activities, which increasingly focused on providing large customers with location-based data. Virtually every time I discussed Swarm with anyone, we shared our mutual confusion and regret over it no longer being called Foursquare.

Over time, my own interest in Foursquare—whoops, I mean Swarm—dwindled. The last time I devoted my full attention to it was in 2019, when I interviewed Foursquare cofounder Dennis Crowley for an article about the company’s evolution over its first decade. Then I went back to ignoring the app until last month, when an ad tech company named Infillion acquired Foursquare, putting it back in the news for the first time in years.

The acquisition’s terms were not disclosed. But in a Substack post this week, Crowley bemoaned the transaction, which he said didn’t involve Foursquare equity holders getting any money.

Crowley—who launched a new AirPods-centric location-based app called BeeBot last year—didn’t mention Swarm. Even Infillion’s own press release touched on the app in the vaguest of terms, saying it was “considering a range of paths including open-source models, strategic partnerships, and opportunities that best support [the Swarm community].”

As long as I was thinking about Swarm again, I decided to, well, check in on it. Many still-extant social networks of yore, such as MySpace and Friendster, have only the most tenuous of connections to their past incarnations. Swarm isn’t like that. While the app hasn’t gone through any dramatic expansion in functionality or ambition, Foursquare has been reasonably diligent about updating it with new features. For instance, it got a map-based Explore tab last April.

Still, I was expecting Swarm to feel hollowed out. After all, market intelligence firm Sensor Tower says it has only 94,000 monthly active users, having lost an average of 38% of its users each year since 2018. That’s about how many followers I once had on Twitter. It hardly sounded like enough people to power an entire social network.

Swarm’s nifty, new-ish map view reminded me that I once checked in from three foreign countries.

Logging in to the app for the first time in eons, I discovered that I had almost 1,300 unanswered friend requests, largely from people I don’t actually know. I would not be shocked if almost all of them abandoned the app years ago. Obvious signs that activity has tapered off included the fact that the user-submitted tips I saw for businesses I checked into tended to date from the distant past. Users also seem to have fallen out of the habit of commenting on each other’s check-ins.

Here’s what surprised me, though: Swarm may be dinky, but it hasn’t lost all its charm. People I friended years ago are still checking in every day. When I resumed doing that myself, some of them noticed. It felt like a reunion.

Why have Swarm diehards stuck around all this time? My wife, who never stopped using the app, says she values having a log of where she’s been and what she’s done. I heard something similar from my friend Owen Thomas, the managing editor of the San Francisco Business Times. He’s a Swarm power user, with 82,850 check-ins as I write this.

Back in the day, Owen told me, he was skeptical of Foursquare. Then he read that Digg cofounder Kevin Rose used the app to record his life story. “That spoke to me,” he says. “Rather than using Foursquare to broadcast to other people, it had an inherent usefulness to me.” He uses an app called Timehop to revisit his past check-ins.

My first Foursquare check-in, from 2009. (It’s wrong about the theater having since permanently closed.)

Somewhere along the way, Owen developed the habit of liking all of his 428 Swarm friends’ check-ins. “I can’t really explain why,” he says. “Except it’s become kind of an in-joke that means something like ‘Thank you for still using Swarm after all these years, and by the way, we’re both still alive, and how great is that?’”

Even with Owen doling out likes with abandon, I find them to be tiny moments of humanity on an internet that’s increasingly being drained of it. Most social networks juice engagement by pelting you with content. On Swarm, there just isn’t that much going on, giving it a placid, calming effect.

Foursquare recently added the ability to connect Swarm to ChatGPT and Claude, but as far as I can tell, there’s no AI slop on the platform. Unlike Instagram, it isn’t bulging with lush photos of users trying too hard to let the world know they’re living their best lives. Actually, I haven’t seen many photos of people at all. Shots of meals, retail environments, and airport terminals predominate. They’re mundane but reassuringly real.

Swarm’s tagline on Twitter/X is “Remember everywhere,” hinting that it’s leaned into the life-logging aspect that my wife and Owen have embraced. With the user base having cratered, it’s no longer a reliable way to determine what your friends are up to right now. But longtime users can always turn to it to jog memories of their own activities from 5, 10, or 15 years ago.

I found scrolling back to the days when I used the app a lot to be an unexpectedly emotional experience, full of reminders that life goes on. My ancient check-ins included defunct businesses I miss, such as Virgin America (my favorite airline) and Fry’s Electronics (my favorite big-box retailer). Today, Swarm informed me that it was the 15th anniversary of me checking in at a brunch place my wife and I liked at the time. Later, we were put off by its prices and the hassle of finding parking. I just learned that it closed in 2018.

As much as I am enjoying reacquainting myself with Swarm, I am wary of getting too attached to it. When I asked Infillion EVP of Marketing Calynn Krieger for an update on the app’s future, she said the company is formulating plans but wants to share them internally before making a public statement. Absent more details, I take its talk of open-source models and partnerships as suggesting it may not want to shoulder full responsibility for the app forever.

If my current return turns out to be a final fling, I’m okay with it. For now, both Swarm and I are still alive—and how great is that?

Are you still on Swarm? I could use more friends, so add me. You’ve been reading Plugged In, Fast Company’s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on fastcompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky, Mastodon, and Threads, and you can follow Plugged In on Flipboard.

More top tech stories on Fast Company ‘You Can See Everything’ perfectly captures what a decade of post-truth has done to us
The three-hour documentary about Elizabeth Holmes gives those of us exhausted by this era of unreality an avatar and champion in Nathan Fielder. Read More →

Why you should worry less about big data centers and more about big data
Google’s $10 million bid for Spirit Airlines’ archive underscores a larger issue: Big AI labs have the money to license or buy massive amounts of business and consumer data. Read More →

The next Xbox could cost $1,800. Here’s why
With Microsoft and Sony facing soaring component costs, the next generation of gaming hardware could come with unprecedented price tags. Read More →

This simple tool fixes Google Calendar’s most frustrating flaw
Turn messy plans into magic calendars that anyone can access—without all the usual headaches. Read More →

AI companies say they want to give billions away. Will they actually do it?

3 weeks 4 days ago

Some of Silicon Valley’s biggest names are digging deep into their pockets. Elon Musk has promised to give every child in America a share in SpaceX next year. That would amount to roughly 72 million shares worth $10.5 billion at recent prices. It’s a generous offer, although each child would receive about $146, a tiny fraction of Musk’s fortune, which has recently topped $1 trillion.

And Musk isn’t alone. Anthropic, the AI company behind Claude, recently unveiled plans to set aside 5% of its shares for public benefit. Based on its most recent funding valuation, that stake would be worth roughly $48 billion. Rival OpenAI has also pledged billions toward charitable causes.

The sums are eye-watering. But turning those promises into money that actually reaches the public is more complicated.

“The traditional approach is one that requires a substantial financial commitment upfront,” says Brian Mittendorf, an accounting professor at Ohio State University, “not just a stated intention to put money aside.”

How the pledges would work

Anthropic’s proposal still requires shareholder approval. If approved, it would transfer shares into a company-controlled limited liability company (LLC), which could donate to charities or distribute money directly to people whose livelihoods have been disrupted by AI. Its board could also allocate another 2% of shares each time the company’s valuation increases by 50%. Including pledges from founders and employees, Anthropic estimates that roughly a fifth of its stock is being dedicated to public benefit.

OpenAI is taking a different approach. Its nonprofit foundation owns a stake in the ChatGPT maker that was valued at around $130 billion when the company restructured last year. The foundation has committed $25 billion toward curing diseases and improving AI resilience, with plans to invest at least $1 billion over a year across those and other causes. It has also awarded $100 million in grants to community organizations through its People-First AI Fund, including $50 million announced this month. The foundation aims to disburse that latest round by year’s end.

Musk’s plan is different still. Reports suggest the SpaceX shares could be distributed through Trump Accounts, the children’s investment accounts established by the Trump administration. Recipients would own stock rather than receive cash, although Musk has yet to explain how the giveaway would work.

A promise isn’t a donation

Despite the unanswered questions, Jay Ritter, director of the IPO Initiative at the University of Florida, sees reason for optimism. “I think we should take them completely seriously,” he says, pointing to a long history of wealthy entrepreneurs funding charitable causes, from John D. Rockefeller to Bill Gates to MacKenzie Scott.

But a pledge isn’t necessarily the same as a donation. Anthropic’s proposed structure, for instance, would allow the company to retain control over when and how its money is distributed, with fewer disclosure requirements than a conventional charity. And while OpenAI has already begun awarding grants, the bulk of its philanthropic ambitions remain commitments rather than money distributed to beneficiaries.

Who benefits from the generosity?

Bella DeVaan, associate director of the Charity Reform Initiative at the Institute for Policy Studies, believes the timing of such promises deserves scrutiny. “I think that we should just always ask who benefits? Who benefits at the timing of the gift?” she says.

She points to her organization’s research into the Giving Pledge, established in 2010 to encourage billionaires to donate at least half their wealth during their lifetimes or after death. The 32 original American signatories who remained billionaires in 2025 had collectively grown 283% wealthier, or 166% after adjusting for inflation. Of 22 deceased pledgers, just eight had fulfilled their commitments.

For DeVaan, those numbers make accountability especially important. “By what mechanisms will a pledger be held accountable to actually fulfill their pledge?” she asks.

Silicon Valley’s latest pledges could ultimately produce substantial benefits, and OpenAI has already begun awarding grants. But the enormous sums being announced also raise questions about how much control these companies are willing to surrender, how quickly the money will be distributed, and what will hold them to their commitments.

These 2 free compression tools protect your privacy

3 weeks 5 days ago

Sometimes you just need to make a file smaller, quality be darned.

Maybe it’s your kid’s school play performance that probably doesn’t need to be preserved in pristine 4K resolution. Or perhaps it’s a PDF that won’t fit within your email provider’s attachment size limits.

Whatever the case, it’s worth bookmarking a couple of free tools that can squish your files down without uploading them to some untrustworthy server somewhere or making any other needless compromises.

This tip originally appeared in the free Cool Tools newsletter from The Intelligence. Get the next issue in your inbox and get ready to discover all sorts of awesome tech treasures!

Quick and easy file compression

For a simple way to make your files smaller, check out ​vidshift.io​ (for videos) and ​nanofile.io​ (for everything else).

➜ Vidshift and Nanofile are free online tools for compressing files, both from the same developer.

⌚ It takes just a minute to start uploading and compressing, though the actual compression times depend on the size of your files.

Both tools are free with no sign-ups or ads, and the compression all happens entirely offline.

1️⃣ Let’s start by looking at ​Nanofile​, which works for images, audio files, and documents. Open up the site, then either use the “Browse your device” link or drag a file directly onto the page. Nanofile will automatically recognize your file type and offer some options to make it smaller.

Nanofile works for images, audio files, and documents.

With MP3s, for instance, Nanofile can lower the bit rate, mix down from stereo to mono, and strip out cover art. For PDFs, you’ll see a quality slider and a way to compress any included images. There’s also a handy comparison for audio files and images, so you can look for any discernible differences. When you’re done, just click the Download button to grab your compressed file.

2️⃣ As for ​Vidshift​, while it’s billed primarily as a way to convert between video file types, it also has a compression tool to make your videos smaller​.

☝️ Make sure to visit ​Vidshift’s Compressor page​, not the main page—which is for converting between video formats. The latter tool can wind up making your videos larger.

After choosing a video, select a target size from one of the presets or use the “Custom” button to choose how small the video should be. (You can see the current size of the video in megabytes underneath the file name at the top.)

Vidshift is incredibly effective at compressing videos, specifically.

The smaller you go, the more you’ll notice blocky compression artifacts in your videos. Using the resolution settings can make your video smaller with less obvious blockiness, though the video won’t look as crisp, especially when viewed on larger screens.

As long as you’re making adjustments, Vidshift has some other useful video editing tools worth noting, including a trimmer​, cropper​, and ​GIF converter.

Note that while both Vidshift and Nanofile require an internet connection to load, the actual file processing happens entirely on your own local device. I confirmed this by adding and modifying files in both tools while my computer was disconnected from the internet.

🎁 One more bonus tip: The same developer also makes a tool for converting text between uppercase and lowercase​, for instance if you type out a rant in ALL CAPS and then decide to tone it down a little. Just don’t expect the quieter prose to make your documents any smaller.

  • Vidshift​ and ​Nanofile​ are both available in any browser, but they work best on desktop devices.
  • Both sites are free to use, with no registration and no usage limits.
  • The sites use Google Analytics to collect aggregate data on things like how visitors arrived at the site and the compression options people choose, but the developer says they have no access to your content and do not share any data for marketing purposes.

Treat yourself to all sorts of brain-boosting goodies like this with the free Cool Tools newsletter—starting with an instant introduction to an incredible audio app that’ll tune up your days in truly delightful ways.

AI stocks slip and global markets retreat as oil hits 5-month high

3 weeks 5 days ago

Drops for stocks in the artificial-intelligence industry are dragging markets down worldwide on Thursday, while oil prices hit their highest levels since May on worries about the war with Iran.

The S&P 500 sank 1.3% following even sharper losses in Asia, whose markets are full of AI and other technology stocks. The Nasdaq composite was down 2.4%, as of 12:56 p.m. ET, but stocks outside of tech did better, and the Dow Jones Industrial Average was down a more modest 40 points, or 0.1%.

AI-related stocks have been shaky on worries that their boom may be under threat because of how expensive it’s become to borrow money to build data centers and buy more AI chips. Concerns also remain that stock prices simply shot too high in their yearslong frenzy, even though companies in the industry continue to report growth.

Lam Research, whose equipment and services help make semiconductors, reported gains in profit and revenue for the summer that were slightly stronger than analysts expected, for example. Usually, that helps boost a stock’s price, but Lam Research’s tumbled 11.1%.

It gave a range for its forecast for profit this quarter whose midpoint was slightly below analysts’ expectations. Its stock also came into the day with a 52.4% surge for the year so far, nearly quadruple the already healthy gain for the overall S&P 500, leaving plenty of room for potential disappointment.

Other big tech stocks were unsettled following their own quarterly reports that included at least some encouraging nuggets.

Meta Platforms, which is pushing its Muse AI agent, lost 7.4% after reporting stronger revenue for the latest quarter than analysts expected, though its profit fell short. The company behind Facebook and Instagram also said it’s creating a new class of stock to raise money to pay for AI investments and other expenses.

Tesla slipped 2.1% after flipping between gains and losses in the morning. It reported a stronger profit for the summer than analysts expected, though its revenue was weaker than forecast.

The global tide against AI stocks knocked Nvidia down 4.3%. Because it’s the largest stock on Wall Street by total value, the chip giant was the heaviest weight on the S&P 500.

CoreWeave, which has borrowed money to buy Nvidia’s AI chips that it rents out to customers over the cloud, has been particularly shaky in recent weeks because of worries about its debt. It dropped 9.6%, though it’s swung between wide gains and losses of roughly 4% within each of the last two days.

The weakness for AI and other tech stocks began in Asia, where South Korea’s Kospi tumbled 3.5%, Hong Kong’s Hang Seng dropped 3%, and Japan’s Nikkei 225 sank 2.7%.

Also weighing on global stock markets was a jump in oil prices.

The price for a barrel of Brent crude oil rose 2.9% to $111.99 after touching its highest level in five months. It’s at the high end of the $97 to $112 range it’s been yo-yoing through in October on uncertainty about when the war with Iran will allow the global energy industry to return to normal.

In Europe, France’s CAC 40 stock index fell to one of the continent’s sharpest losses with a drop of 1.4%. It has fewer tech stocks than Asian or U.S. indexes, but it was hurt by another jump in yields for French government bonds.

Yields have been generally climbing there because of worries about the government’s big debt load and strained budget. Violent protests across France have raised pressure on the government to increase spending, which could add further to its debt.

Worries about big government debt are weighing on almost every developed economy in the world, including the United States. Such concerns have pushed yields in the U.S. bond market to their highest levels in decades, along with high inflation, signals of a still-strong U.S. economy and other factors.

But U.S. yields nevertheless fell sharply Thursday in a signal that investors were buying U.S. government bonds in search of something safer to own as AI stocks tumbled.

The yield on the 10-year Treasury fell to 4.98% from 5.13% late Wednesday and from more than 5.30% earlier this month. That’s a major move for the bond market, though the 10-year yield remains well above its 3.97% level from before the war with Iran.

Thursday’s drop in yields came even as another report suggested the U.S. economy may be doing better than economists expected. It said fewer U.S. workers applied for unemployment benefits last week, which could signal that layoffs remain low across the country.

That solid economy has helped U.S. companies broadly to make bigger profits, which is what propelled the U.S. stock market to a record high earlier this month despite the many worries tugging at it. The majority of stocks within the S&P 500 rose on Thursday, only to get drowned out by the drops for more influential tech stocks.

Procter & Gamble, Lockheed Martin, and CSX were among the latest companies to join the list of those topping profit expectations for the summer, and gains for their stocks helped limit the market’s losses on Thursday.

CEO Shailesh Jejurikar said Procter & Gamble expects “geopolitical and economic challenges” to continue, but it stood by its forecasts for growth in revenue, profit, and other measures for the full year. The stock price for the company behind the Crest, Olay and Oral-B brands rose 6.3%.

—By Stan Choe, AP business writer

AP Business Writer Elaine Kurtenbach contributed to this report.

Why you should worry less about big data centers and more about big data

3 weeks 5 days ago

Welcome to AI Decoded, Fast Company‘s weekly newsletter that breaks down the most important news in the world of AI. I’m Mark Sullivan, a senior writer at Fast Company, covering emerging tech, AI, and tech policy.

Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at sullivan@fastcompany.com, and follow me on X @thesullivan.

Google’s Spirit Airlines deal could be just the start

While data centers remain a major issue in the midterm elections, another aspect of the AI industry’s expansion is drawing far less attention: the enormous amounts of data tech companies are amassing to train their models.

Consider Google’s offer to pay $10 million for Spirit Airlines’ business data archive, which is being sold as part of the airline’s bankruptcy proceedings. The archive contains roughly 100 million employee emails, 500 million Microsoft Teams messages, and 30 million lines of source code, along with pricing curves, flight operations data, and crew schedules. Google wants the information to help train its AI models.

Early AI chatbots were trained primarily on information scraped from the internet. But as AI companies develop agents capable of handling complex business tasks, they’re increasingly seeking out real-world business data that can help their models understand how organizations operate and solve problems. That has fueled a growing market for licensed data, from Reddit posts and newspaper archives to healthcare research and other industry-specific information.

Companies have been licensing their data for years, but acquiring the internal records of a bankrupt business represents a different opportunity. Rather than simply teaching Google’s models about the airline industry, Spirit’s archive could help them learn how employees communicate, collaborate, and navigate complex problems. An AI agent might, for example, learn from the steps workers took to resolve a particular issue, then apply a similar approach to an entirely different problem.

And Google may not be the last AI company to pursue such a deal. “I think you probably see it more and more now as people realize how valuable a data set is,” says Alex Kimata, a data privacy attorney at Holland & Hart. While $10 million is a substantial sum in bankruptcy proceedings, it’s relatively little for a company like Google, particularly if buying an entire archive proves cheaper than licensing comparable information.

For now, the fate of Spirit’s data rests with a federal bankruptcy court in the Southern District of New York. Google emerged as the winning bidder at auction, with the AI recruiting company Mercor named the backup bidder at $7.5 million. Another company, Micro1, subsequently submitted a $12.5 million bid in an attempt to wrest the deal from Google ahead of the court’s final decision on November 6.

The proposed purchase raises significant privacy concerns. Spirit’s broader data holdings include millions of travelers’ names and credit card numbers, as well as information about thousands of employees. Google says personally identifiable information will be excluded from the sale and has agreed to remove certain categories of sensitive data and hire a third party to anonymize the archive.

But the flight attendants union has objected, arguing that even after identifying information is stripped away, employees and customers could potentially be identified again.

That’s a legitimate concern, says Matthew Kohel, a privacy attorney and partner at Saul Ewing. Even when names and other identifiers are removed, information from separate sources, such as airline loyalty programs or in-flight purchases, can sometimes be combined to reconstruct someone’s identity. “People that are motivated have ways of recreating individuals’ identities across different data sets,” he says.

Google faces a difficult balancing act. It needs the information to be sufficiently anonymized to satisfy privacy requirements without stripping away so much detail that the data loses its value for AI training.

“It’s just really hard to do that and meet the legal standard of actually being de-identified,” Kimata says. He considers the union’s skepticism warranted, given the difficulty of ensuring that individuals cannot be identified from the remaining information.

And privacy is just part of a broader issue. AI companies have a massive and growing hunger for data, and with much of the publicly available web already used for training, they’re increasingly turning to more specialized sources. In the age of AI agents, the more valuable, and scarcer, resource is process data; agents can train on data about how a worker or individual works through a complex task, step by step, and how they use connections to apps and tools. Spirit’s archive is one example of how companies can acquire that information at scale.

The terms of service for consumer AI are shifting to match a similar training need. Anthropic, for example, moved to training on consumer chats by default unless users opt out. AI companies are now competing to provide a personal agent for everyone (see Meta’s Muse, OpenAI’s Dots, and Anthropic’s Duet). As these agents gain utility and user trust, they may become receptacles of huge amounts of personal data, and with that data, they may be able to do more and more of the user’s thinking for them.

Of course, all that business and personal data flowing into a handful of closed, opaque frontier models puts an enormous amount of power into the hands of the companies that own them. And the risks inherent in that may go beyond privacy. The AI companies that train and tune models that millions of people use to draft, decide, and research may end up quietly setting what counts as a reasonable answer, with little public input.

Perverse business incentives might create new risks: An assistant that knows a person’s fears and habits, paired with an ad- or engagement-driven business model, might become a persuasion engine that seems neutral but subtly steers users toward choices that are profitable for the AI company. Over time, heavy reliance on the same few systems could homogenize how people frame and solve problems; meanwhile, users’ natural faculties for doing those things might erode.

And the Spirit deal may be just the beginning. For AI training, buying a data store describing the deep internal workings of a whole company just makes sense. Saul Ewing’s Kohel notes that for $10 million Google is basically buying “a company in a box,” and for cheap: To a company like Google, $10 million is little more than “a rounding error on their balance sheet.” And the data it acquires could help shape products that millions of people eventually rely on.

OpenAI says two of its agents disabled safety monitoring

OpenAI disclosed Monday that agents weakened misalignment monitoring while carrying out two internal coding tasks. On October 3, a GPT-6.1 Sol agent repairing an evaluation pipeline disabled automatic alerts to OpenAI’s security team. Three days later, a GPT-6 Astra agent migrating a program removed its monitoring integration, then falsely told the user it had not. “In both cases, the agents considered whether monitoring safeguards should be preserved, but interpreted routine coding instructions as permission to weaken them,” OpenAI wrote. The company says it still lacks a satisfying explanation for the incidents.

Anthropic launches Claude Duet, an agent that messages you first

Anthropic on Wednesday launched a research preview of Claude Duet, an AI agent app powered by Claude Opus 5.5 and aimed at professionals and small business owners. Duet is proactive, Anthropic says, reaching out to users rather than waiting for instructions. It delivers a morning brief of priorities pulled from connected apps, then sends messages throughout the day when something needs attention. Users can also assign tasks of varying complexity, from scheduling meetings and building dashboards to creating financial forecasts, prototyping, and developing web apps. Duet is available on iOS, Android, and claude.ai, and is included in Pro and Max plans.

More AI coverage from Fast Company:

Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company today.

How Google, Amazon, and Meta are trying to win over communities pushing back against data centers

3 weeks 5 days ago

With host communities across the U.S. turning against their gargantuan data centers, America’s technology giants are promising to take fewer taxpayer handouts, keep fewer secrets or shower more money onto the communities willing to allow the server warehouses.

The new raft of promises from the world’s richest companies comes as anti-data center sentiment has become a potent issue in the midterm elections, with control of Congress and 36 governors’ offices on the line.

Despite backing from President Donald Trump, the data centers powering the artificial intelligence boom have become politically toxic. Community opposition that blocks municipal permits has become a bottleneck, and tech giants could face even more challenges if voters next month elect more public officials who are skeptical of data centers.

The tech industry says it has gotten the message. In addition to ramping up financial incentives, officials with Google and Meta said in interviews with The Associated Press they are trying to address mistrust by being more transparent with communities where they would like to build.

Companies are also making pledges to reduce their impact on the environment and offering assurances that data centers will boost local property values — all while maintaining AI is a positive force for the American economy, despite its skeptics.

Tech giants pledge more transparency

Devi Thomas, Google’s global head of community development and engagement, told AP that she wouldn’t call resistance to data centers a “backlash,” but that “there are often community questions to be addressed.”

“Early public consultation is increasingly a very important part of the process. We have done as much as we can to make sure communities are involved early on,” Thomas said.

Earlier this month, Amazon rolled out its “Amazon Data Center Commitment,” including pledges to improve the environmental rating of its backup diesel generators and disclose more about its data centers’ energy and water use. It also said — as have Microsoft and Google — that it had stopped requiring government officials to sign non-disclosure agreements.

The use of non-disclosure agreements has meant people in host communities may know little or nothing about a project until it shows up on a municipal board’s voting agenda. Besides that, tech companies often have hidden their involvement in a project behind a code name and developers who do the legwork with local governments, critics and other developers say.

That practice “raises people’s suspicion as to what’s truly happening,” said Anita Verma-Lallian, a developer with data center projects in Arizona.

Meta is always evaluating how well it shares information with communities, said Lindsay Elin, Meta’s vice president of public policy and external affairs. “As part of that work, we are reviewing our use of non-disclosure agreements,” she said in a statement to the AP.

In a report this month that sharply criticized data center giants including Amazon, Google, Meta and Microsoft, several Senate Democrats said companies have used non-disclosure agreements to freeze out the public. The report also said the companies aggressively seek state and local tax breaks and do not cover the entire cost of the electric-grid buildout needed for energy-hungry data centers.

Spending ramps up in host communities, along with skepticism

The dominant players in the industry — Amazon, Google, Meta, and Microsoft — project they’ll spend more than $600 billion this year combined on capital projects, primarily data center construction in the U.S.

Even as the companies prepare to spend more to entice communities — payments that some critics call “bribes” — Trump urged them in a social media post Wednesday to pay people and communities a “dividend” if they “want to make their Data Centers popular.”

“MAKE DEALS, and get them built!!! GREAT FOR THE U.S.A.,” Trump wrote.

Meta and Amazon said recently they will each spend $1 billion on financial incentives to host communities in the coming years, on top of millions they already spend to support schools, charities and job training. Google says its investments in local programs in data center communities will reach $1 billion by the end of 2027.

Microsoft has pointed to a couple local tax breaks that it will forego. And Google recently said for all of its new data center campuses, the company will pay full real property taxes on the buildings and land.

The companies have known since last fall that community opposition was becoming a serious problem.

Data center demand began driving up electric bills last year across a 13-state electricity grid stretching from Illinois to New Jersey and Maryland, angering ratepayers. The outcry led to promises from the tech giants to pay the cost of the power infrastructure to feed their data centers, while Microsoft rolled out its five-point “Community-First AI Infrastructure” plan.

Data centers’ critics aren’t persuaded.

Sam Burleigh, who helped start Concerned Citizens of Montour County in rural northern Pennsylvania to fight a potential Amazon-backed data center project, said the company’s commitments are “propaganda.”

“What I see is them realizing they have more resistance than they ever dreamt of. And so now they have to find some way, some method to break through that resistance,” Burleigh said.

Kathryn Hoffman, executive director of the Minnesota Center for Environmental Advocacy, which has sued over two prospective Google data centers, said she sees Google doing more public relations outreach, but she doesn’t see it doing more to disclose critical details about its plans.

Political stakes are rising

In an Aug. 18 memo, the Senate Republicans’ national campaign arm crystallized the midterm election pressure on the tech giants.

It warned the tech sector that it must “fix the toxic brand of an entire segment of the economy” or risk long-term fallout in elections. That included addressing “who benefits, who pays, and why a community should want one.”

Drew Nations, a developer in North Carolina, said developers need to aggressively address quality-of-life complaints from neighbors. That includes shielding neighbors from noise and data centers from public view so they aren’t an eyesore, and they need to forego “corporate welfare” such as the lucrative sales tax exemptions that nearly 40 states offer.

Those tax breaks leave regular people thinking, “‘So you’re saying that this multi-multi-billion if not trillion-dollar company gets tax breaks and I don’t?'” Nations said.
Meta, Google, Microsoft and Amazon collectively reported more than $300 billion in net income in their most recent fiscal year.

Spending to promote data centers has been on the rise. The Data Center Coalition, a trade association whose members include the tech giants, said it has spent more than $10 million running pro-data center ads in 10 states on TV and elsewhere. Data center giant QTS and a group that doesn’t disclose its donors, Build American AI, also began ad campaigns this year.

Meanwhile, angry residents continue to flood once-sleepy municipal board meetings, pressuring officials to reject data center proposals.

More municipalities are passing data center moratoriums and Data Center Watch, a project of AI security consultancy 10a Labs, said it saw sustained local opposition to dozens of projects in the April-to-June quarter, its latest reporting period.

Balaji Tammabattula, chief operating officer of BaRupOn, which develops data center campuses, said the pressure of the midterms has helped developers understand how crucial community support is to a project’s success.

“It’s impossible to build without them,” he said.

Follow Marc Levy at http://x.com/timelywriter

—Marc Levy and Kaitlyn Huamani, Associated Press

Checked
15 minutes 10 seconds ago
Fast Company Technology
Fast Company inspires a new breed of innovative and creative thought leaders who are actively inventing the future of business.
Subscribe to Fast Company Technology feed