Trump Mandates Pre-Release Government Review for Advanced AI Models

Under a new executive order signed Tuesday, AI developers must grant the U.S. government access to their most powerful models up to 30 days before public release. This voluntary framework aims to mitigate national security risks while navigating fierce industry pressure to keep regulatory hurdles to a minimum.

The policy targets high-stakes players like OpenAI, Google, and Anthropic, establishing a mid-point between the 14 days initially sought by tech firms and the 90-day window originally proposed by the White House. The directive emerged following concerns over Anthropic’s Mythos model, which remains withheld due to its potential to exploit vulnerabilities in critical infrastructure, including banking and hospital systems.

Beyond pre-release screening, the administration is establishing an AI cybersecurity clearinghouse. This joint effort between the Treasury, the National Security Agency, and CISA intends to identify and patch software weaknesses in collaboration with private sector operators. While industry leaders like Sam Altman and Google’s Kent Walker have expressed support for the balance struck by the order, critics remain skeptical. Anthony Aguirre of the Future of Life Institute argued that voluntary measures are insufficient, insisting the government requires authority to actively block releases that pose unacceptable security threats. The rollout of this policy follows internal friction within the administration, fueled by venture capitalist David Sacks, who warned that such oversight risks stifling American innovation in the global race against China.