Trump Administration Negotiates Access to Anthropic’s Powerful Mythos AI Model

Anthropic's powerful new artificial intelligence model, Mythos, has become the focus of intense negotiations between the company and the Trump administration. According to Bloomberg, White House Chief of Staff Susie Wiles met with Anthropic CEO Dario Amodei on Friday, with the White House characterizing the discussion as "productive and constructive." The meeting signals that the administration is seeking wider access to Mythos, despite significant security concerns surrounding the model's capabilities.

Mythos represents a significant technological breakthrough but also poses considerable risks. According to CSIS Wadhwani AI Center senior adviser Gregory Allen, the model is being examined for its potential military applications, particularly in the context of the ongoing Iran conflict. The AI system has demonstrated advanced capabilities, including the ability to identify and exploit thousands of zero-day vulnerabilities across every major operating system and web browser. In testing, Mythos uncovered a 27-year-old bug in critical internet software and a 16-year-old flaw in video code, showcasing its exceptional reasoning and coding skills.

Anthropic has chosen to severely limit public access to Mythos due to these dual-use concerns. Rather than releasing the model widely, the company launched Project Glasswing, granting early access to major technology partners including Apple, Amazon, Microsoft, Google, Nvidia, and several cybersecurity firms. According to reporting on the matter, these organizations will use the new AI model to hunt for vulnerabilities in their own software and share their findings to inform future safety guardrails. However, Anthropic's internal red team demonstrated that the model could be weaponized to chain together exploits against every major operating system and web browser, leading the company to forgo a general public release.

The situation has become more complicated by the Pentagon's recent actions. A federal appeals court unanimously ruled against Anthropic in its legal dispute with the Defense Department, denying the company's motion to stay the Pentagon's ban on Anthropic's goods and services during ongoing litigation. The court determined that while Anthropic would likely experience harm from the ban, a balancing test favored the government, particularly given the current military conflict in Iran.

The White House meeting reflects a broader tension: Mythos appears too strategically important for the U.S. government to abandon entirely, yet its capabilities present undeniable cybersecurity risks. Further testing of the model is expected to continue in the coming months as stakeholders attempt to establish protocols for safe deployment while maximizing the model's potential benefits for national defense and cybersecurity applications.