Microsoft CEO Satya Nadella says AI labs are quietly stealing their customers’ know-how

• Satya Nadella warns enterprises they’re paying for AI twice.
• Thinking Machines Lab ships its first model. 
• Google DeepMind loses another researcher.
• And, Anthropic’s CEO pours $1 million into a PAC pushing AI regulation.

Are Big Tech companies turning against frontier AI labs?

In the past few weeks, several executives and VCs have voiced concerns that the AI labs are gaining outsized access to their customers’ most sensitive business data.

This week, Microsoft CEO Satya Nadella weighed in. He published a blog post warning that companies using AI models like OpenAI’s and Anthropic’s are “paying twice”: once for tokens that do the work, and again by handing over the proprietary knowledge needed to make the models useful.

“Models learn from ‘exhaust,’” he wrote, “the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong. Every correction is distilled into institutional know-how.”

That’s a problem, Nadella argues, because enterprises are unsuspectingly handing the labs the ability to one day compete with them. While AI labs freely train on the open internet, he added, they restrict enterprises from doing the same to their models.

His fix is for companies to retain ownership of their own data and build “proprietary learning environments” in the cloud, paired with “orchestration layers” that let them switch between AI providers instead of getting locked into one.

Does Microsoft happen to sell exactly this kind of proprietary cloud environment and orchestration layer? Funny you should ask…

So sure, Nadella is being a bit self-serving. But it’s nonetheless a somewhat surprising argument to come from the CEO of a company that has poured billions into both OpenAI and Anthropic. But Nadella isn’t alone.

On CNBC earlier this month, Alex Karp, CEO of Palantir Technologies, argued frontier AI labs are quietly extracting the very thing that makes a business valuable—proprietary data, processes, and competitive edge—while charging premium token prices for tools he says often don’t deliver commensurate value.

Karp said enterprise clients are privately “livid” about their AI vendors, arguing frontier labs are more interested in “tokenmaxxing” than solving real business problems. Palantir followed up with a nine-point “AI sovereignty” manifesto declaring that “data retention is your treasure.”

Even from outside the Big Companies, people are piling on.

On his All In podcast, former White House AI and Crypto Czar, David Sacks, called Karp “exactly right” and argued that OpenAI and Anthropic have formed a duopoly that leaves enterprises with too little leverage over their own data and infrastructure.

Karp’s argument, like Nadella’s, is good for his own business. Palantir’s core pitch is that its Foundry platform lets enterprises use any model without exposing their “alpha”—the proprietary data, processes, and know-how that make up a company’s competitive edge.

The open-source argument

All of this rhetoric may also push companies toward open-weight models—models whose underlying parameters are published publicly, letting anyone download, run, and modify them—and away from close-sourced models such as GPT-5 or Claude. These open-source models can offer enterprises more control and transparency over their data, often at a fraction of the cost of frontier models.

Open-weight models—the most advanced of which are coming from Chinese-based companies—are also closing the performance gap with the frontier and seem to be gaining in popularity with users.

Vercel, for example, says open models now account for 29% of traffic through its AI gateway. Amazon’s CTO, Werner Vogels, also recently told me that he was seeing companies shift from frontier model providers to open source options, both in an attempt to reign in cost and gain more transparency over the technology they are embedding into their companies.

The U.S. government’s recent decision to block access to Anthropic’s Fable 5 models has also made companies consider diversifying away from a single AI vendor. After all, if companies are relying on AI models to run crucial workloads, they want to be sure no one can pull the plug overnight.

With that, here’s more AI news.

Beatrice Nolan
beatrice.nolan@fortune.com
@beafreyanolan

FORTUNE ON AI

Meet the AI employee that convinced Sequoia to invest $45 million in Sable — by Lily Mae Lazarus

Commentary: Genesys CEO: We can see firsthand how AI is changing — not replacing — work — by Tony Bates

Why IBM just suffered its worst stock crash of all time—and what it says about the market’s two bubbles — by Nick Lichtenberg

AI IN THE NEWS

xAI sues Grok user over sexualized deepfakes. Elon Musk’s xAI has sued a South Carolina man who was arrested in February on child sexual abuse material charges. The company alleges he misused Grok to generate sexually exploitative images of minors and non-consensual sexual imagery of adults, violating the company’s terms of service. It’s one of the first cases of an AI company suing its own user over generated content. xAI’s complaint says it suspended more than 52,000 accounts and made over 73,000 reports to the National Center for Missing & Exploited Children in 2026, leading to at least 244 arrests. The company is seeking monetary damages and a permanent ban on Harwood’s use of Grok. The suit follows sustained scrutiny of xAI, including a lawsuit from Baltimore, over Grok’s alleged role in producing non-consensual sexualized deepfakes, including of children, earlier this year. Read more in Reuters.

OpenAI staffers bankroll a rival to their boss’s super PAC. Rank-and-file OpenAI employees have donated more than $215,000 to Guardrails Alliance, a super PAC pushing for stricter AI regulation. It’s intended as a counterweight to Leading the Future, the pro-industry PAC backed by OpenAI president Greg Brockman, who has donated over $100 million. Seven current employees and one former staffer have given money, including a $200,000 donation from research engineer Juan Felipe Cerón Uribe. Other donors include safety researcher Gabriel Wu and alignment researchers Julie Steele and Jason Wolfe. OpenAI says Brockman’s involvement with Leading the Future is personal, not corporate. Guardrails Alliance, which has raised $5 million towards a $15 million goal, joins Anthropic-backed Public First Action in opposing the group. Read more in Wired.

Thinking Machines Lab ships its first model. Mira Murati’s Thinking Machines Lab has released its debut model, Inkling, an open-weight system trained to handle audio and video alongside text. The company says Inkling doesn’t top the benchmarks but performs well across a range of tasks, including reasoning and coding. Notably, Thinking Machines used Inkling to fine-tune itself, and noticed the model’s reasoning traces grew more concise over time. The release is a bid to compete with the Chinese labs currently leading open-weight AI, and aligns with the company’s stated view that a handful of firms shouldn’t control AI. Thinking Machines launched in February 2025 with one of the largest seed rounds in history. Since then, however, it’s lost several of its executives to OpenAI. Read more in Wired.

Google DeepMind researcher quits over military deal. AI safety researcher Alex Turner has resigned from Google DeepMind, saying he spent months trying to stop the company from signing a military deal with the Pentagon. He says the deal violates Google’s 2018 pledge against developing or supporting lethal autonomous weapons. In a post on X, Turner said he raised the issue directly with CEO Demis Hassabis, who directed him to two senior policy staffers, but the proposal went unanswered until the deal was signed. He also pushed back on Hassabis’s public claim that “nothing’s changed” about DeepMind’s principles, noting Hassabis co-authored the post that removed the weapons prohibition from Google’s AI Principles. Turner said he saw three honest options: explain the change, formally renounce the pledge, or leave. While he acknowledged that Anthropic held its red lines when tested on similar issues, most other companies have fallen in line. Read the full post on X.

Hassabis, Altman, and Amodei converge on calls for regulation. Demis Hassabis, Sam Altman, and Dario Amodei have each published proposals in the last five weeks arguing that the frontier of AI needs regulation. It’s a rare moment of public alignment for the CEOs of Google DeepMind, OpenAI, and Anthropic. All three want independent testing of frontier models before release, some governing body that can certify compliance and restrict access to the riskiest systems, and a U.S.-led international framework rather than a patchwork of state or international rules. The three diverge slightly on how to go about it, however. Amodei wants an FAA-style agency with the power to block a model’s release, Hassabis proposes a FINRA-style industry body starting with voluntary reviews, and Altman is pushing an IAEA-style international forum that uses market access as leverage. Critics have also noted that established labs may benefit from this. Many of the larger labs already have the legal and compliance infrastructure to handle a review process, raising fears the rules could entrench incumbents over startups and open-source developers. Read more on Axios.

EYE ON AI NUMBERS

$1 million

That’s how much of his own money Dario Amodei just put behind Anthropic’s push for stronger regulation. Federal Election Commission filings released Wednesday night show the CEO personally donated $1 million in May to Public First, the super PAC pushing for mandatory AI safety rules—his largest political gift on record. It came weeks before Public First–aligned groups spent roughly $12 million trying, unsuccessfully, to elect New York’s Alex Bores, the state lawmaker whose AI safety law made him the top target for Leading the Future, the rival network backed by OpenAI president Greg Brockman. Several more Anthropic staffers added over $2 million in combined income around the same time; the filings even revealed a quarter-million-dollar gift from someone at Google DeepMind, plus a smaller one from an OpenAI employee.

The amount of money being poured into political campaign groups is also a sign of how central AI has become to U.S. politics at the moment. AI companies are also reckoning with a rising tide of AI backlash, something that will likely spill over into the midterms. Last month, a Reuters poll found just 33% of Americans think the rapid data center buildout is mostly a good thing, while 77% worry it’ll push up their electricity bills, a concern that cuts evenly across party lines. Read more from Politico here.

AI CALENDAR

Aug. 4-6: Ai4 2026, Las Vegas.

Nov. 16-17: Fortune 500 Innovation Forum, Detroit. Apply here to attend.

Dec. 6-12: Neural Information Processing Systems (Neurips) conference. Sydney, Australia.

Dec. 7-8: Fortune Brainstorm AI, San Francisco. Apply here to attend.

Watch: Can AI smell you?

For decades, scent remained computing’s final analog frontier. Now, Osmo has accomplished what many thought impossible: digitizing smell. By training AI to decode the relationship between molecular structures and human perception, Osmo has built the world’s first olfactory intelligence platform—transforming scent into searchable, programmable data. While poised to reshape the flavor and fragrance industry, its implications extend far beyond perfume, opening new possibilities across health, food, technology, and human-computer interaction. Watch the video here.