Anthropic CEO Dario Amodei to Meet White House Over Access to Mythos AI Model
Key Points
- Anthropic CEO Dario Amodei will meet White House chief of staff Susie Wiles on Friday to discuss federal access to Mythos.
- Mythos can identify and exploit thousands of zero‑day vulnerabilities, succeeding on the first attempt in over 83% of cases.
- Anthropic created Project Glasswing, a controlled‑access program that currently serves about 40 vetted organizations.
- The Pentagon labeled Anthropic a national‑security supply‑chain risk after the firm refused to lift safety restrictions.
- Multiple U.S. agencies, including Treasury and CISA, are seeking defensive use of Mythos despite the blacklist.
- Global regulators in the UK, Canada and the U.S. are coordinating responses to the model’s security implications.
- A potential deal could restore Anthropic’s eligibility for government contracts while keeping safety safeguards on autonomous weapons.
Anthropic chief executive Dario Amodei is set to sit down with White House chief of staff Susie Wiles on Friday to negotiate federal access to Mythos, the company’s frontier AI system that can discover and exploit thousands of zero‑day vulnerabilities. The talks come after the Pentagon blacklisted Anthropic for refusing to lift safety guards on its models, even as Treasury, intelligence agencies and the Cybersecurity and Infrastructure Security Agency press for controlled use of the technology. The meeting could shape the future of AI‑driven cybersecurity policy in the United States and abroad.
Anthropic’s chief executive Dario Amodei will meet White House chief of staff Susie Wiles on Friday, marking the most concrete step yet toward easing the standoff between the AI firm and the Pentagon. The discussion centers on Mythos, a general‑purpose artificial‑intelligence model that surprised its creators by pinpointing and exploiting thousands of previously unknown zero‑day flaws across major operating systems and browsers. When tasked with building working exploits, Mythos succeeded on the first try in more than 83% of cases and completed a 32‑step corporate network attack simulation from start to finish.
Instead of releasing Mythos publicly, Anthropic rolled out Project Glasswing, a tightly controlled access program that supplies the model to roughly 40 vetted organizations, including Amazon Web Services, Apple, Google, Microsoft, Nvidia, Cisco, CrowdStrike, JPMorgan Chase and Palo Alto Networks. The firm pledged up to $100 million in usage credits and $4 million in donations to open‑source security groups, positioning the model as a defensive tool for finding and fixing critical software vulnerabilities before adversaries can weaponize them.
The clash with the Department of Defense began in February when Defense Secretary Pete Hegseth demanded unrestricted access to Anthropic’s models for any lawful purpose, including autonomous weapons and domestic surveillance. Amodei’s refusal led Hegseth to label Anthropic a national‑security supply‑chain risk—a designation previously reserved for companies linked to foreign adversaries—effectively blacklisting the firm from defense contracts. Anthropic sued the administration in March, alleging illegal retaliation. A federal judge initially blocked the blacklist, but an appeals court reversed that decision on April 8, leaving the company barred from DoD work while litigation continues.
Paradoxically, the same government now seeks Mythos for its own cybersecurity needs. The Treasury Department wants the model to hunt for flaws in its systems, while parts of the intelligence community and CISA are already testing it. The White House Office of Management and Budget is drafting safeguards to let federal agencies use a controlled version. Reports indicate Anthropic hired consultants with ties to the former administration to smooth negotiations, and Friday’s meeting is expected to lay the groundwork for a possible deal.
Industry leaders have underscored Mythos’s significance. JPMorgan Chase CEO Jamie Dimon warned that the model “reveals a lot more vulnerabilities” than traditional tools. The UK’s AI Security Institute described a preview as “substantially more capable at cyber offence than any model previously assessed.” The Council on Foreign Relations called the development an “inflection point for AI and global security.” If the technology falls into hostile hands, the consequences could be catastrophic; if it remains with defenders, it could close security gaps that have persisted for decades.
Globally, regulators are scrambling to understand the implications. The Bank of England’s governor flagged Mythos as a cybersecurity risk, prompting an emergency briefing with the CEOs of the UK’s eight largest banks and key financial‑infrastructure providers. Anthropic plans to extend Project Glasswing to select British banks within days and is expanding its London office to 800 staff. Canadian finance officials have labeled the model an “unknown unknown” in discussions at IMF meetings, while regulators in the United States, United Kingdom and Canada coordinate on assessment frameworks.
A potential resolution would restore Anthropic’s eligibility for government contracts, grant controlled access to Mythos for defensive purposes, and see the Pentagon drop the supply‑chain risk label. In return, the company might agree to a review process for limited military use cases that do not violate its safety principles on autonomous weapons and mass surveillance. Both sides have incentives to compromise: Anthropic needs a clear path to work with the broader U.S. government, and the administration needs a powerful tool to protect critical infrastructure. The outcome of Friday’s meeting could set a precedent for how AI safety, commercial interests and national security intersect in the era of advanced artificial intelligence.