Back in March, I wrote about why I cancelled my OpenAI subscription: the company had quietly traded its founding mission for military contracts and a for-profit future. Two headlines from the past couple of weeks caught my attention, and I think they are worth flagging together, because neither one is a scandal by itself, but they continue the same trajectory.
A Proposal, Not (Yet) a Transfer
On July 2, 2026, multiple outlets reported that OpenAI had pitched the idea of giving the U.S. government a 5% equity stake in the company. I want to be precise about what that means, because it is easy to read a headline like that and assume the deal is done. It is not. This is a concept under early discussion, reportedly floated by Sam Altman in talks with Commerce Secretary Howard Lutnick and Treasury Secretary Scott Bessent. Nothing has been signed, and it is unclear whether it will happen at all.
The structure being discussed resembles the Alaska Permanent Fund, the sovereign wealth vehicle that invests the state's oil revenue and pays residents an annual dividend. OpenAI's version envisions similar stakes from other major AI companies, including Anthropic, Google, and Meta, flowing into a shared fund. Reporting on the proposal is explicit that "it's unclear whether any other company would agree" to something like this, and there is no indication any of them have.
So why mention a proposal that might go nowhere? Because the fact that OpenAI is the one floating it says something. This is a company whose original mission statement promised development "unconstrained by a need to generate financial return," a phrase that, as I noted in March, has since been quietly dropped. A company still holding that founding line seems unlikely to be the one pitching Washington on a sovereign wealth arrangement. Whether or not this particular proposal survives, it is another data point in the same direction: OpenAI increasingly treats its equity and its relationship with the federal government as tools to manage political risk, not as lines it is reluctant to cross.
Another Safety Leader Heads for the Door
The second story is a personnel change that, on its own, might not warrant a blog post. Johannes Heidecke, OpenAI's head of safety systems since 2024 (he joined the company in 2021 and succeeded Lilian Weng in the role), is leaving as part of a reorganization. Safety and research are being merged under Mia Glaese, who takes on a new title of VP of Research and Safety. Saachi Jain steps in as interim head of safety systems, reporting to Glaese.
Chief Research Officer Mark Chen framed the change as integration rather than reduction: "It's important that our safety work is integrated with frontier-model development, with an earlier and more direct role in shaping key model, product and launch decisions." That is a reasonable-sounding rationale, and it may even be genuine. But it is worth reading against the pattern: OpenAI's Superalignment team, dedicated specifically to long-term AI safety research, was dissolved in 2024. Its Mission Alignment team was disbanded in February of this year. Now the standalone head-of-safety role is being absorbed into a research leadership structure, timed to a period the company itself describes as training models "at a much faster cadence" with shorter release cycles.
Each of these changes has an internal logic. Taken together, they describe an organization where dedicated, independent safety leadership keeps shrinking relative to the pace of model development.
Why Flag This At All
I want to be fair here: companies reorganize. Executives leave. Proposals get floated and quietly shelved all the time. If I wrote a post every time one of those things happened at OpenAI, this blog would be nothing else. What makes these two worth pairing is that they land on the same axis as everything I wrote about in March: a company moving closer to government and military entanglement while its independent safety infrastructure gets thinner. The military pivot, the Pentagon contract, the mission statement rewrites, and now a pitch to hand Washington equity and a reorganization that folds safety into the research org under faster release pressure. It is not one event. It is a direction.
Where I Land
My recommendation from March has not changed. I still use and recommend Anthropic's Claude for the nonprofits and educators I work with, in part because Anthropic's public commitments on safety have been tested under real pressure and held. That is not a permanent endorsement; I will keep watching both companies, and I will write again if something changes the picture. For now, this is simply the next chapter of the same story.
If you have questions about which AI tools make sense for your organization, or want a second opinion on a vendor's track record, use the contact form to start a conversation.
