White House races to finalize voluntary frontier-model review framework ahead of August 1 deadline

A 60-day clock set by President Trump's June 2, 2026 executive order, formally Executive Order 14409 on AI cybersecurity and frontier model security, expires August 1, when the White House is expected to formalize a voluntary framework giving federal agencies early access to review covered frontier models before their public release. Under the plan, developers including OpenAI, Anthropic, Google, Microsoft, and Amazon would give the government access to a qualifying model for up to 30 days before it reaches other trusted partners, a window intended to let agencies assess national security and safety risks ahead of launch. Meta is notably absent from the negotiations, since the review mechanism is built around pre-release access to a model, and Meta's Llama models are released as open weights that cannot be meaningfully restricted once they are public. In practice, the government has already shown it does not need a formal framework to intervene: the Commerce Department used separate export control authority, which predates the executive order by roughly eight years, to suspend global access to Anthropic's Claude Fable 5 and Mythos 5 models and to restrict OpenAI's models on two separate occasions in the past two months. That pattern is why observers describe the arrangement as voluntary on paper but closer to mandatory in practice, since the same agencies negotiating the framework already have separate legal tools they are willing to use if a company declines to cooperate. The framework's finalization will be closely watched as an early test of how the administration intends to balance AI innovation with national security oversight without formal binding regulation.