Federal agencies were supposed to complete a classified process for testing advanced AI models’ cyber capabilities, but no public announcement has confirmed delivery and talks with major developers were still continuing in late July.
The Aug. 1 deadline in President Donald Trump’s AI executive order has passed without a public announcement that the U.S. government completed its framework for evaluating advanced AI models before release. The order, signed on June 2, 2026, directed agencies including the NSA, CISA, Treasury and NIST to build a classified benchmarking process for stress-testing the cyber capabilities of frontier AI systems and determining whether they qualify as a "covered frontier model," a designation that would bring added government oversight. The order also set up a voluntary process under which developers could give the government up to 30 days of pre-release access to models. Negotiations with major AI companies including OpenAI, Anthropic, Google, Microsoft and Amazon were still underway in late July, underscoring how compressed the timeline was for a classified interagency effort. Meta was not part of those talks, reflecting the tension between the administration’s controlled-access approach and the company’s open-source strategy, which centers on releasing models such as Llama for broad public use. The delay leaves unresolved how the administration will treat open-weight AI systems, an issue with implications for decentralized AI projects that rely on openly available models.