Today marks the first formal deadline in US history for government oversight of frontier AI model releases. A 60-day clock set by President Trump's June 2 executive order — EO 14409, 'Promoting Advanced Artificial Intelligence Innovation and Security' — expires on August 1, when the White House is expected to formalize a voluntary pre-release review window for the most capable AI systems.

The negotiations involve OpenAI, Anthropic, Google, Microsoft and Amazon — but not Meta, whose open-weight Llama models sit structurally outside any lab-level access control. The framework was drafted in close consultation with the labs it will govern, a provenance the Financial Times first reported in early July.

What the deadline formalizes is a process already tested twice in practice. In June, the Commerce Department suspended global access to Anthropic's Claude Fable 5 and Mythos 5 after a jailbreak exposed advanced cyber-offense capabilities, using export control authority that predates the order by eight years — legal analysts at Mayer Brown called it a possible first application of the Export Administration Regulations to an AI model itself. Weeks later, the White House asked OpenAI to limit the June 26 launch of GPT-5.6 Sol to government-vetted partners; it became broadly available on July 9 after a 12-day gated preview.

The order's 'voluntary' language is real but incomplete: nothing in EO 14409 creates mandatory licensing or preclearance, yet the Commerce Department's Bureau of Industry and Security retains independent statutory authority under the Export Control Reform Act of 2018 to impose controls on technologies 'essential to national security.' Brad Carson of Public First described the current state bluntly: 'You have an ad hoc, personalized, opaque, possibly lawless approach.'

The operational heart of the framework is classified. The NSA director holds sole authority to designate which models count as 'covered frontier models,' based on a benchmarking process run with CISA and Treasury that assesses autonomous cyber capabilities — the order's language targets models able to 'find and exploit weaknesses in software on its own.' The Five Eyes alliance warned in June that AI models capable of launching major cyberattacks are 'months — not years — away.'

Whatever the August 1 announcement contains, the practical precedent is already set: in seven weeks, the government suspended one frontier model globally and gated another, without a published threshold. Labs now have every incentive to trade predictability for a voluntary commitment.

Sources