Trump Administration Reviews Frontier AI Model Policies After Anthropic Mythos Decision

The Trump administration is examining whether to establish a more formal oversight process for certain advanced AI systems after Anthropic chose not to release a model called Mythos that can independently discover software vulnerabilities. According to the Washington Post, the company determined that the risks of making the model widely available outweighed any potential benefits.

Vice President JD Vance addressed the issue in a recent call with leaders from Microsoft, Alphabet, OpenAI and Anthropic, citing the potential for such tools to enable cyberattacks on small-town banks, hospitals and water plants.

The discussions reflect a shift from the administration’s earlier emphasis on limiting regulatory burdens. Last year at a Paris AI summit, Vance cautioned that overregulation could harm the industry. In December 2025 he issued an order targeting state AI laws.

Officials have begun comparing potential new review mechanisms to the FDA’s drug approval process, according to the Washington Post. The emerging framework under consideration would likely apply only to a small number of cutting-edge systems with exceptional capabilities in areas such as vulnerability detection rather than the broader AI ecosystem.

Anthropic is not the first company to withhold a powerful model from public release. The Washington Post reported that the decision on Mythos has accelerated internal administration debates about how to address AI systems capable of autonomously identifying and potentially exploiting weaknesses in existing software.

Participants on the Vance call included representatives from major technology firms actively developing frontier AI systems. The conversation focused on the intersection of advanced AI capabilities and national cybersecurity, particularly risks to localized critical infrastructure with limited cybersecurity resources.