The White House just told OpenAI and Anthropic that US intelligence gets first look at new models — while one of the sharpest investors in Silicon Valley says those models won't be worth the compute bills anyway.
The Summary
- The White House Office of the National Cyber Director asked OpenAI and Anthropic to hold new models from UK testers until US government review, forcing companies to choose between UK access and Trump administration favor
- Scott Wilson, CIO of Washington University's endowment who called SpaceX early, now says frontier labs are "not worth the liabilities they signed up for" as cheaper Chinese models close the gap
- The UK's AI Security Institute previously enjoyed privileged early access to frontier models, positioning this as a geopolitical shift in AI testing hierarchy
- Wilson passed on early OpenAI investment despite schools like University of Michigan turning $20M into $2B+, betting the unit economics never pencil
The Signal
The Trump White House is asserting control over AI model releases in a way that reconfigures the global testing order. OpenAI and Anthropic must now choose: give US intelligence agencies first crack at new models, or maintain their relationship with the UK's AI Security Institute. The senior administration official told POLITICO this applies to "every new frontier model that comes out." The stated reason is security. The real reason is who gets to set the rules.
The UK built one of the most well-resourced government testing agencies in the world. They had early access. That access is now conditional on US approval. This isn't about safety testing timelines. It's about which government decides what "safe enough to deploy" means for American companies building the most powerful AI systems on Earth.
"The White House wants the US to review the models and take steps to make sure US systems are secure before the models are shared with US partners."
Meanwhile, one of the investors who saw SpaceX coming thinks the entire frontier lab model is broken. Scott Wilson manages Washington University's endowment. His SpaceX bet created billions in returns. He had the chance to invest early in OpenAI. He passed because he didn't believe the company would ever make money. His view hasn't softened. It's hardened.
Wilson's thesis: OpenAI and Anthropic committed to trillion-dollar compute spending just as Chinese competitors ship models at a fraction of the cost. "There's going to be a ton of free alternatives," he said. The infrastructure bills come due regardless of whether the models maintain their lead. OpenAI's early investors like University of Michigan saw $20M turn into $2B+, but Wilson argues those markups don't reflect sustainable unit economics.
The counterpoint matters here: Vinod Khosla, an early OpenAI investor, told Business Insider he strongly disagrees. Khosla believes the massive infrastructure investments are exactly what will make frontier labs hard to beat. This isn't a consensus bear case. It's a specific bet that the cost curve breaks against US labs faster than quality advantages compound.
The timing is sharp. The White House is asserting sovereign control over model releases at the exact moment smart money is questioning whether those models justify their cost structure. If Wilson is right, the US government is fighting for first access to assets that are about to get commoditized. If Khosla is right, the US just secured pole position in the most important technology race of the century.
The Implication
Watch what happens to UK-US AI cooperation over the next six months. If the UK pushes back hard, this becomes a wedge issue in the special relationship. If they accept it quietly, expect other Five Eyes countries to accept similar terms. The testing hierarchy matters because it determines which government's red lines become the global standard.
For OpenAI and Anthropic, this is the cost of staying onshore. Chinese labs don't answer to the White House. They also don't get Pentagon contracts or intelligence community work. The frontier labs are now explicitly choosing which sovereign they serve first. That choice has a price, and Wilson thinks it's higher than the market realizes.