Anthropic and Mythos: Trump Considers Drastic AI Regulation

The Trump administration is considering implementing strict government oversight on artificial intelligence (AI) models before they hit the market. This initiative, revealed by the New York Times and anonymous sources within the government, is largely motivated by the Mythos model from Anthropic. This model has been deemed too dangerous by the company itself, which has drawn the attention of federal agencies.

The Mythos Model in the Spotlight

The Mythos model from Anthropic is capable of identifying thousands of critical software vulnerabilities. This capability has piqued the interest of the NSA, which has used it to assess flaws in Microsoft systems within the administration. However, some agencies remain without access to this tool. Analysts question the danger posed by Mythos, pointing out that other, less expensive models offer comparable performance.

Anthropic has also refused to modify its models to make them compatible with controversial applications such as autonomous weapons and mass surveillance. This stance has led the Pentagon to view the startup as a risk to the supply chain, resulting in the cancellation of a $200 million contract. A federal judge described this decision as "Orwellian," illustrating the tensions between Anthropic and the government.

Towards a British Model of Oversight

The oversight project envisioned by the Trump administration would draw inspiration from the British AI Security Institute. A working group, including tech leaders and government officials such as the NSA, the national cyber director, and the national intelligence director, would be tasked with evaluating AI models before and after their deployment. This system would allow the government to access models prior to their market release without delaying their launch.

Representatives from Anthropic, Google, and OpenAI have reportedly already been informed of these plans. This project comes as David Sacks, who was in charge of AI policy under Trump, left his position in March, leaving the chief of staff and the Treasury Secretary to manage AI policy. The White House has characterized this information as "speculation," clarifying that any official announcement would come from Trump himself.

The issue of AI model security is now being addressed at the highest levels of the U.S. executive branch. While Anthropic had sought regulation under Biden, it may see its wishes fulfilled under Trump, albeit without controlling the terms. Meanwhile, open source models continue to develop outside any regulatory framework.

In January 2025, the Trump administration had repealed the Biden executive order on AI security, making this turnaround particularly notable.