A voluntary government review of an AI model, meaning a safety check that a company can decline to participate in but still chose to do, has already been applied to OpenAI's next major release. Sam Altman, OpenAI's chief executive, called the Trump administration's review of the forthcoming Astra model a "productive process" and said closer engagement with government safety offices will become more important as AI systems grow more capable.
OpenAI is preparing to release Astra, a model the company has designated at its "critical" cyber capability level. That designation, the first of its kind for OpenAI, signals the company's own judgment that the system requires stronger safeguards during development and before it reaches the public.
What the government's framework actually covers
The Trump administration has built a voluntary framework for reviewing certain advanced AI models before release. Under that framework, government officials could receive access for up to 30 days. The White House has no plans to make the framework public. That means most companies outside the process, and the general public, cannot see the rules being applied to some of the most capable AI systems currently being built.
Altman acknowledged the review was optional. "But we of course did it," he told Axios. He added that as models reach what he called "a quite significant level of capability," engagement with safety institutes in the United States, the United Kingdom, and elsewhere will become more important.
The positions being staked out at the G20
Altman traveled to the G20 Innovation Ministerial to appear alongside Commerce Secretary Howard Lutnick before an audience of commerce and technology ministers. He described OpenAI as "the pragmatic centrists" and said people, rather than AI, should run the future.
Nvidia CEO Jensen Huang also spoke at the summit. His message to foreign ministers was direct: the worst outcome is choosing not to take advantage of AI and being left behind. That framing drew visible agreement from ministers in the room.
Altman pushed back against both extremes. He said what he called "doomerism," pursued for the sake of gaining power or enforcing a worldview, is harmful. Blind optimism that ignores risk is equally bad, he said. Both OpenAI and Anthropic have drawn criticism that their support for some forms of AI regulation could benefit larger, established labs while raising barriers for smaller competitors.
On Congress, Altman said he expected lawmakers to establish a basic framework for advanced AI after his first testimony in 2023. That has not happened. He attributed the inaction to the pace of the technology itself: people have a hard time knowing how to regulate the right things without slowing innovation, and getting Congress to act on large issues appears, in his telling, to be very difficult.
Altman said OpenAI has "much, much, much more capable models coming soon" and called on governments to start thinking now about what their societal impact could be.