Anthropic CEO to White House as AI access debate heats up

A pivotal meeting is set for this Friday between Anthropic CEO Dario Amodei and White House Chief of Staff Susie Wiles, centered on negotiating government access to the company’s groundbreaking Mythos AI model. This high-stakes discussion aims to resolve a major impasse with the Pentagon, which blacklisted the AI firm earlier this year. The urgency stems from the model’s unprecedented ability to discover and weaponize thousands of zero-day vulnerabilities, a capability that has drawn intense interest from agencies like the Treasury Department, CISA, and the U.S. intelligence community.

The conflict originated in February when Defense Secretary Pete Hegseth demanded Anthropic remove its safety guardrails to allow for potential use in autonomous weapons and surveillance. Amodei’s refusal led to the company being designated a national security supply-chain risk, a move typically reserved for foreign adversaries, and barred from Defense contracts. Anthropic responded with federal lawsuits alleging illegal retaliation. While a court initially blocked the blacklisting, an appeals court reversed that decision earlier this month, sustaining the standoff.

This creates a stark paradox: the same administration that penalized Anthropic now actively seeks its most powerful creation. Project Glasswing, Anthropic’s controlled access program, already provides Mythos to roughly forty vetted entities, including major tech firms and financial institutions like JPMorgan Chase, for defensive cybersecurity research. The model is not a security product, but a general-purpose AI that, during testing, proved capable of executing complete, multi-step network intrusion simulations, successfully developing exploits on its first attempt in over 83% of cases.

The global response to Mythos has been swift and concerned. The Bank of England’s governor recently highlighted it as a major cybersecurity risk. The UK’s AI Security Institute, which has an evaluation partnership with Anthropic, assessed a preview and found it “substantially more capable at cyber offence than any model previously assessed.” Canadian officials have discussed its implications at IMF meetings, and UK financial regulators are convening emergency briefings with the nation’s largest banks.

This international attention adds a geopolitical dimension to the White House negotiations. Anthropic is preparing to provide Mythos access to select British banks and is significantly expanding its London office. This raises the possibility that America’s closest ally could gain operational access to this critical tool before the U. S. government itself, a scenario that undoubtedly motivates the administration to find a compromise.

A potential deal would likely see the Pentagon withdraw its supply-chain risk designation, restoring Anthropic’s eligibility for contracts. In return, the company would provide Mythos access for defensive cybersecurity purposes across the government while maintaining its core restrictions on autonomous weapons and mass surveillance. Both sides have compelling reasons to settle. The blacklisting undermines Anthropic’s credibility with enterprise clients, while the government cannot afford to ignore a tool of such profound defensive and offensive significance.

The meeting underscores a central tension in AI governance. The capabilities of frontier models are advancing faster than regulatory frameworks can adapt. Anthropic’s trajectory illustrates this perfectly, a company rewarded by the market for its safety-first approach yet penalized by the state for those same principles, only to be courted again when its technology became indispensable. The debate over Mythos, a tool so powerful that both restricting and releasing it are defensible positions, is no longer theoretical. It is now being negotiated in the highest corridors of power.