In the space of about ten days, OpenAI lost its head of safety systems, got sued by Apple over stolen hardware secrets, watched the federal government gate the launch of its newest models, and offered Washington a $42 billion slice of itself.
Read those as four separate bad headlines and you miss the point, because they are all the same story: a company that no longer wins on the product trying to win on proximity to power instead.
The Ledger
Start with the receipts, because the pile is genuinely unusual for a company this size.
Apple sued OpenAI on July 10, alleging that a departing iPhone engineer walked out with confidential hardware files and a company MacBook he never returned. CNN Business reported that the complaint names OpenAI, the engineer, and hardware chief Tang Tan, who ran iPhone and Apple Watch design before he left, along with io Products, the Jony Ive company OpenAI bought last year. OpenAI told TechCrunch on July 14 that it was “not aware of any evidence that this complaint has merit.” Apple has since put roughly 40 former employees now working at OpenAI on notice to preserve documents, which is not the behavior of a plaintiff planning to settle quietly.
Days earlier, Johannes Heidecke, the company’s head of safety systems, told staff he was leaving. As Engadget noted, his exit followed a reorganization that folds OpenAI’s safety teams into research, reporting up through a newly expanded VP of research and safety role.
And the newest models did not ship on OpenAI’s schedule. TechCrunch reported in late June that the GPT-5.6 rollout was staggered at the request of the US government under a voluntary framework granting federal reviewers up to 30 days of early access to frontier models. OpenAI said publicly that it did not think this kind of process should become the default. It complied anyway.
The Thing Nobody at OpenAI Wants to Say Out Loud
Here is the structural cause underneath all of it. OpenAI is not the default anymore.
The independent analysts at Epoch AI have been tracking the revenue crossover, and by their read the gap between OpenAI’s self-reported run rate and its closest rival’s has not just closed, it has inverted. Fortune framed the same shift on July 2 as Sam Altman seeking a new world order for AI precisely because OpenAI is losing ground to Google and Anthropic. Consumer subscriptions, the business OpenAI built, turn out to be a worse moat than enterprise contracts, which is the business it did not build.
That reframes every other item on the list. You poach aggressively from Apple’s hardware org when you need a device business you cannot grow organically. You collapse safety into research when safety is a function that slows shipping and you can no longer afford to ship slowly. Neither of those is a scandal on its own. Together, in the same fortnight, they are a company under pressure making the trades that companies under pressure make.
Selling Washington a Stake in Your Own Survival
Which brings us to the $42 billion.
CNBC reported on July 2 that OpenAI has floated giving the US government a 5% equity stake, worth roughly $42 billion against a valuation near $840 billion, and that the idea has been raised with President Trump, Commerce Secretary Howard Lutnick, and Treasury Secretary Scott Bessent. Altman has called on Google, Meta, and Anthropic to do the same. None have taken him up on it. We covered the mechanics of that proposed federal stake in OpenAI when it first surfaced.
The pitch is that this makes AI labs accountable to the public. It does the opposite, and the reason is not complicated. Accountability means the state can impose costs on you when you cause harm. Equity means the state books a loss when you take one. Those are opposite incentives wearing the same suit.
A regulator holding 5% of the company it regulates is not a watchdog with teeth. It is an investor with a position to protect, sitting in the same building as the people who decide whether your model is safe to release, whether your competitor gets an export license, whether your antitrust exposure is worth pursuing. The Commerce Department is already the body vetting frontier models before public launch. Handing that department’s government a nine-figure quarterly interest in one vendor’s stock price is a conflict you cannot disclose your way out of.
Notice also what the offer is not. It is not board seats, not audit rights, not binding safety commitments, not a seat for labor or for the communities absorbing the data-center buildout. It is a check, and checks do not come with subpoena power.
The Timing Is the Tell
Sequence matters here. This proposal did not arrive during OpenAI’s dominant stretch, when it had the leverage to set terms and the market position to refuse them. It arrived in the same month the company lost its revenue lead, lost its safety chief, lost control of its launch calendar to federal reviewers, and picked up a trade-secrets fight with the most litigious hardware company on earth. Our earlier reporting on Apple’s trade-secrets suit against OpenAI laid out how thin the company’s public denial looked next to the specificity of Apple’s filing.
Companies negotiating from strength do not offer a government 5% of themselves. They lobby, they litigate, they wait out the news cycle. Offering ownership is what you do when you have concluded that the market will not protect your position and something else has to.
The bet is legible enough. If Washington owns a piece of you, Washington needs you to survive. Export controls, procurement, model clearances, and antitrust all start bending toward the balance sheet. That is a rational play for OpenAI’s shareholders and a bad outcome for everyone who has to live with whatever the company ships next, because the entity best positioned to say no acquires a financial reason to say yes.
What to Watch
The number to watch is not OpenAI’s valuation. It is whether any other lab takes Altman up on the offer. He needs company: a single firm with a government stake is a captured vendor, while four firms with government stakes is an industry structure, and industry structures are much harder to unwind than one embarrassing deal.
So far nobody has moved. If that holds through the fall, the $42 billion offer ages into an artifact of a bad month, the moment a company that used to set the terms started asking for protection instead. If it does not hold, the question stops being whether the US government will regulate artificial intelligence and becomes whether it can still afford to.