OpenAI and Anthropic — ordinarily the fiercest rivals in the AI model market — are quietly cooperating in Washington on a federal process that would give US agencies up to 30 days of pre-release access to advanced frontier AI models before they ship to customers or trusted partners.
The collaboration surfaced in late-July industry reporting and comes ahead of an Aug. 1 deadline for federal agencies to define which AI systems qualify as "covered frontier models" — the trigger threshold that would subject the most-advanced systems to mandatory pre-release government evaluation.
Why two rivals are collaborating on policy
OpenAI and Anthropic sell directly competing general-purpose AI systems and battle for the same enterprise contracts, top researchers, compute capacity and investor dollars. And yet on this specific policy question — what should the federal government require of advanced AI models before they ship — the two companies appear to be aligned.
Both firms reportedly want any pre-release access framework to apply industry-wide, not just to developers already voluntarily cooperating with Washington. The practical consequence: rivals like Meta and xAI could be brought under the same review framework as long as their models cross the eventual capability threshold. That shared standard is more attractive to both OpenAI and Anthropic than a fragmented, voluntary-only regime that only the early-movers opted into while the rest of the industry shipped however they chose.
Neither company has publicly released the full text of the proposed framework terms.
The Trump executive order behind the deadline
The 30-day pre-release review process is being grounded in Executive Order 14409, signed by President Donald Trump in June, which directed federal agencies to build a classified benchmarking process for advanced AI models and submit a formal framework by Aug. 1, 2026.
The order sets a voluntary — for now — 30-day window during which developers of certain frontier models would share them with agencies for evaluation before broader distribution. The capability threshold for "covered" status remains undefined; that is the Aug. 1 deliverable agencies are now working toward.
Not all of the AI industry is on the same page
The OpenAI-Anthropic alignment is not industry-wide consensus. Other major AI players have already pushed back against parts of the proposed framework. Nvidia, Meta and Microsoft separately warned against sweeping new controls on open-weight AI models — arguing that open-weight models, by definition, allow anyone to download, modify and re-deploy the underlying weights, and that broadly applied pre-release access rules would be functionally impossible to enforce on open-weight releases.
That pushback matters because many frontier-class models are now released under open-weight licences — even if not always with the same degree of openness — and the practical scope of any pre-release framework depends on whether those models fall in or out.
Implications beyond AI
The policy alignment matters even outside the AI-model markets because frontier-class AI is increasingly intertwined with crypto mining and AI infrastructure — a thesis the public Ionic Digital listing this week underscored with its dual Bitcoin-mining and AI-data-center positioning. The $2.83-billion market-cap debut on Nasdaq cemented a hybrid model that increasingly turns AI infrastructure into its own liquid asset class — meaning AI policy rules increasingly matter to crypto-mining and infrastructure equities.
Watch for the Aug. 1 framework definition as the next milestone. A narrower definition of "covered" models would leave OpenAI, Anthropic and Google's frontier models inside the regime and Meta's open-weight Llama family largely outside it; a broader definition would extend the pre-release evaluation requirement across most large foundation-model releases in the US.
