White House Moves to Vet AI Models After Anthropic Halts Mythos Release Over Security Vulnerabilities

Anthropic has voluntarily postponed the release of its latest AI model, Mythos, after internal testing revealed the system could identify thousands of vulnerabilities in operating systems and web browsers—capabilities that could be weaponized by cybercriminals or hostile foreign agents to compromise critical infrastructure, national economies, and military systems worldwide. The company has restricted access to approximately 50 companies and organizations while the risks are assessed, marking one of the most significant self-imposed delays in AI deployment to date.

The Trump administration is now developing a federal review process that would evaluate the safety of powerful AI models before approving their public release, according to a report in The New York Times. The initiative represents a notable departure from the administration's generally anti-regulatory stance toward industry, signaling that AI capabilities may have crossed a threshold where national security concerns are overriding free-market preferences. The move follows growing recognition that advanced AI systems may pose risks that individual companies cannot adequately assess or contain on their own.

The Mythos case exposes the central tension in AI safety: models designed to find and fix security vulnerabilities can equally be used to exploit them. Anthropic's decision to limit distribution rather than release openly reflects an emerging consensus that the most powerful AI systems may require new governance frameworks—whether through industry self-regulation, federal oversight, or a combination of both. As AI capabilities continue to advance, the question of who decides when a model is safe enough for release—and what standards apply—is becoming one of the most consequential policy debates in technology.