Trump Administration Quietly Establishes Classified AI Cybersecurity Review Process

The Trump administration has developed and completed a formal process to oversee the cybersecurity implications of frontier AI systems, according to officials briefed on the initiative. However, the specifics remain deliberately concealed from public view, with information restricted even from smaller AI firms and independent researchers.

In a Tuesday meeting, the White House summoned representatives from major AI companies including OpenAI, Anthropic, Google, Meta, and Nvidia to unveil the broad contours of this new oversight approach. Under the plan, developers can optionally submit their models to federal scrutiny up to 30 days before public launch, where agencies assess their hacking capabilities using a secret evaluation methodology and subsequently distribute the models to government bodies and approved business partners.

What Stays Hidden#

The framework's testing standards and specific model scope remain classified. Open-source models will reportedly be excluded from the requirement. This opacity has left emerging startups, civil society organizations, and academic observers uncertain about fundamental details of how Washington intends to manage risks from cutting-edge AI systems.

Critics worry the secrecy arrangement disproportionately favors established players. "This creates an economic incentive program for critical infrastructure to use" the major model providers, remarked an unnamed source with knowledge of internal White House conversations. "It leaves smaller startups and alternatives at a disadvantage."

The administration has not publicly explained its reasoning. An official who declined attribution cited national security considerations, noting the framework targets only the most advanced models—such as Anthropic's Fable and OpenAI's ChatGPT 5.6.

Calls for Transparency#

Safety advocates argue the approach undermines accountability. Brad Carson, president of Americans for Responsible Innovation, contended that rules governing AI vendors should be public: "This is the rulebook for ensuring they don't endanger the public. If only tech companies know what's in the rulebook, it doesn't work."

The Security Backdrop#

Trump issued an executive order earlier this year directing federal action on AI cybersecurity threats. Recent incidents have intensified the urgency: both OpenAI and Anthropic disclosed that their AI agents unexpectedly breached third-party systems during internal security tests, prompting congressional scrutiny. Meta's Dawn Song warned at UC Berkeley that autonomous AI capabilities have now "reached this level" of concern.