White House Considers Pre-release Vetting of AI Models

According to The New York Times, the Trump administration is discussing an executive order that would establish an "AI working group" and a formal government review process for new AI models before they are released to the public. The New York Times reports White House officials last week briefed executives from Anthropic, Google, and OpenAI on these plans. Tom's Hardware and other outlets report the proposed system could grant government agencies early access to models without necessarily blocking releases, and that agencies under consideration to oversee reviews include the NSA, the Office of the National Cyber Director, and the Office of the Director of National Intelligence, as described in media coverage of unnamed U.S. officials.

Reporting by The Next Web, Tom's Guide, and other outlets documents that Anthropic rolled out a limited-release cybersecurity model, Mythos, via Project Glasswing and that company materials or demonstrations claimed the model autonomously discovered numerous zero-day vulnerabilities across major operating systems, including specific long-standing bugs cited in coverage. Forbes and Bloomberg reporting, summarized by Forbes, say an Office of Management and Budget memo discussed protections and possible controlled access that would allow federal agencies to use a version of Mythos.

Public coverage frames Mythos as a frontier-capability model focused on cybersecurity tasks, with reporting describing autonomous vulnerability discovery and exploit generation in evaluated demonstrations. Industry reporting highlights two technical vectors that make such models notable for policy: models that can autonomously explore attack surfaces at scale, and models whose deployment raises operational capacity and access-control questions when usage is limited to few organizations.

Reporting frames the White House deliberations as a policy reversal from the administration's earlier deregulatory posture, citing The New York Times account that the administration previously revoked a Biden-era AI executive order. Media accounts link the shift in part to concerns raised by the Mythos demonstrations and to internal White House personnel changes referenced in Tom's Hardware and The New York Times. For practitioners, a formal pre-release review regime or mandated government access to certain models would change release cadence, compliance needs, and vendor-government interaction models, particularly for models claiming dual-use capabilities in cybersecurity.

Observers should track whether an executive order is formally issued, the charter and membership of any AI working group, and whether the review process includes binding certification, simply advisory assessments, or guaranteed early-access mechanisms. Reported indicators to follow include any OMB guidance implementing controlled-access provisions for specific models, statements from companies briefed by the White House, and technical disclosure practices around frontier-capability evaluations.