Trump AI order sets voluntary pre-release checks for frontier models

Trump’s AI executive order is a course correction from earlier proposals that flirted with mandatory preclearance for powerful models. Instead, it opts for a voluntary 30-day pre-release window, framed around cybersecurity and critical infrastructure, with a classified benchmark to decide which systems count as “covered frontier models.” That keeps the US closer to a partnership model with labs rather than a licensing regime akin to nuclear export control.

Strategically, this gives leading US labs a clearer regulatory lane: if they engage, they get government help probing models for cyber risk and a stronger argument that they are acting responsibly. If they don’t, they risk political blowback and potentially harsher rules later. The order also creates institutional hooks — an AI cybersecurity clearinghouse, a formal designation process — that future administrations could tighten or expand.

For the AGI race, the impact is ambiguous. On one hand, the absence of mandatory controls reduces friction for US labs trying to move quickly, especially relative to more prescriptive regimes abroad. On the other, serious voluntary collaboration on cyber-capable models could surface vulnerabilities earlier, lower the odds of catastrophic misuse and build the trust needed to sustain aggressive investment. The real question will be how many companies opt into the framework once their most capable models are on the line.