Anthropic Confirms It Briefed Trump Officials On Withheld Mythos Model

Anthropic co-founder Jack Clark said Thursday that the AI firm notified the Trump administration about Mythos, a powerful new model the company is keeping from public release because of serious cybersecurity concerns. Clark, who also leads the company’s public benefit efforts, made the disclosure at the Semafor World Economy summit amid a running legal fight with the Department of Defense that has labeled Anthropic a supply-chain risk.

Company materials and reporting last week described Mythos as having capabilities that worried Anthropic’s own safety teams, prompting the firm to restrict access rather than roll it out broadly. Clark’s acknowledgment follows published reports that White House officials had encouraged several large banks — including JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America and Morgan Stanley — to experiment with the model in private tests. Anthropic has said the model’s cyber-related functions make a public release unwise at this stage.

The briefing revelation arrives against the backdrop of a March lawsuit Anthropic filed after the Defense Department placed the company on a supply-chain risk list. Anthropic and Pentagon officials have clashed over how and whether the military should be able to use the company’s systems, with the company expressing concern about unfettered access for use cases it contends could include intrusive surveillance or autonomous weapons. A separate deal ultimately went to a different vendor.

Clark minimized the label as a narrow procurement disagreement and stressed that the dispute should not prevent conversations with federal agencies about national security. He said Anthropic believes government officials need visibility into advanced systems and that the company wants to establish new, more flexible ways to work with regulators and state actors while protecting sensitive capabilities. Clark also indicated that Anthropic will continue to brief officials about Mythos and future models.

Outside analysts say the episode highlights a tension that has become familiar in tech policy: private companies both seeking government partnerships and resisting government control. An independent AI policy analyst noted that firms frequently share information with regulators to manage risk, but doing so while litigating a federal agency can complicate trust and oversight. The analyst added that asking banks to trial a nonpublic model raises fresh questions about financial-sector risk management and potential systemic exposure.

Clark also addressed labor markets and higher education. He said his team’s economic monitoring has so far identified only limited softness in early postgrad hiring in a few fields, a more cautious assessment than some of the firm’s public warnings about large-scale job displacement. When pressed about advice for students, Clark recommended degrees that emphasize cross-disciplinary thinking and strong analytical skills, arguing that future workers will need to pose the right questions and synthesize knowledge from many domains rather than rely on narrow rote expertise.