Dario Amodei, CEO of Anthropic, is set to meet Susie Wiles, White House Chief of Staff, to discuss access to Mythos, a frontier AI model capable of identifying and exploiting numerous zero-day vulnerabilities across major operating systems and browsers. The meeting is pivotal in resolving Anthropic’s conflict with the Pentagon, which arose after Amodei refused to lift safety restrictions on the AI models. US Treasury, intelligence agencies, and CISA, along with UK financial regulators, are seeking access to Mythos through Anthropic’s Project Glasswing program.
Mythos is not a dedicated cybersecurity product; it’s a general-purpose AI model that can identify thousands of previously undiscovered zero-day vulnerabilities, outperforming both human efforts and automated testing. The model, however, is not publicly available and is provided through Project Glasswing, which includes select organizations tasked with securing software. Anthropic has invested significantly in supporting these efforts.
The Pentagon conflict escalated when Defense Secretary Pete Hegseth demanded unrestricted access to Anthropic’s models for lawful purposes, including potential military applications. Amodei resisted, resulting in Anthropic being blacklisted by the Pentagon. Anthropic filed lawsuits against the Trump administration, contesting the blacklisting. Despite setbacks in court, the company continues to collaborate with other government bodies.
Although Anthropic is blacklisted, the government still seeks access to Mythos for its advanced capabilities. Treasury and intelligence sectors are already testing it, and the White House is finding ways to secure its use by federal agencies. Anthropic has hired consultants to assist in negotiations, with the Wiles meeting aimed at establishing a pathway to collaboration.
Mythos has garnered attention globally; JPMorgan Chase noted its ability to expose cyber threats, while the UK AI Security Institute highlighted its unique offensive capabilities. The Council on Foreign Relations views it as pivotal for AI and global security. The dual nature of Mythos, both protective and potentially destructive, underscores the importance of Anthropic’s controlled access approach.
Anthropic’s solid financial standing strengthens its negotiation position. While it doesn’t rely on Pentagon contracts, it seeks a compromise respecting its safety principles and government collaboration. This meeting is a step towards resolving their conflict while considering global ramifications and security interests.
As Mythos stirs international concern, Anthropic is expanding its reach, offering Mythos access to selected British banks and growing its London operations. The complex geopolitical landscape involves balancing national security needs and ethical deployment of AI technology.
A potential resolution may involve Anthropic regaining government contract eligibility while limiting Mythos use to defensive cybersecurity purposes. Discussions would address military applications while upholding ethical boundaries. The essential discourse over Mythos highlights a broader narrative in April 2026 about AI governance, technological advancements, and institutional adaptation.