White House Replaces Anthropic CEO Dario Amodei in AI Model Talks with Co-founder Tom Brown

The Trump administration has grown frustrated with Anthropic CEO Dario Amodei during negotiations over reinstating the Claude Fable 5 AI model, which was taken offline June 12 via export controls. The White House now prefers dealing with co-founder Tom Brown, who one insider said is not a 'weirdo' like Amodei and can productively engage. Brown and Anthropic's public policy chief Sarah Heck have been leading recent calls with the administration, both at high and working-group levels. The key sticking point remains how to prove Fable 5 is immune to jailbreaks—the NSA found that the restricted Mythos model could have its guardrails bypassed. Independent cybersecurity experts argue that guardrails are only a stopgap, as skilled users and future models will inevitably break them. Commerce Secretary Howard Lutnick is taking a leading role, with a bipartisan group of lawmakers pressing for specific criteria and a timeline to restore access. The White House is also dealing with unrelated controversies, such as the algae blooms in the Lincoln Memorial reflecting pool, but the Anthropic story is a significant flashpoint in AI governance.

The administration's shift in personnel reflects a desire for more straightforward negotiations—something they believe Tom Brown provides. Despite the warmer tone, export controls remain in place, and no timeline has been set for Fable 5's redeployment. A letter from representatives Liccardo, Obernolte, Franklin, and Lieu demanded responses by June 26 on what specific criteria the Commerce Department uses to restore public access. The Bureau of Industry and Security manages these export controls, and Lutnick has been instrumental in addressing jailbreak risks. The talks involve technical staff evaluating proof levels for robustness. Anthropic's spokesperson has not commented, and the White House declined to confirm progress. The uncertainty underscores the broader challenge: balancing innovation with national security as AI models become more powerful and harder to constrain. The outcome could set a precedent for how the U.S. government handles frontier AI models with dual-use capabilities.