Anthropic's Mythos positioning

Anthropic’s new Claude Mythos Preview is becoming a test of whether a company that fought the Pentagon can still win Washington. (theverge.com) Anthropic introduced Mythos Preview on April 7 as a “general-purpose frontier model” and said its strongest early use case is cybersecurity: finding and fixing software flaws before attackers do. The company tied the launch to Project Glasswing, a program with partners including Amazon Web Services, Apple, Cisco, Google, Microsoft, Nvidia, Palo Alto Networks and the Linux Foundation. (anthropic.com) Anthropic’s technical write-up said Mythos Preview had already found “thousands of high-severity vulnerabilities,” including flaws in major operating systems and web browsers, and that it was being released first to a limited set of defenders. Anthropic also published a system card and a separate risk update alongside the launch. (anthropic.com 1) (anthropic.com 2) (anthropic.com 3) The Washington angle comes from what changed after that launch. The Verge reported April 18 that Anthropic had been trying to get back into the U.S. government’s good graces and that Chief Executive Dario Amodei attended a White House meeting on Friday. (theverge.com) Reuters and The New York Times reported that the April 17 White House meeting was the first such discussion after an earlier dispute between Anthropic and the Pentagon over how the company’s models should be used. The Times said U.S. officials viewed Mythos as potentially important for security work. (reuters.com) (nytimes.com) That dispute grew out of a much larger defense relationship. A Congressional Research Service brief said the Department of Defense announced contracts in July 2025 worth up to $200 million each for Anthropic, Google, OpenAI and xAI, and said Anthropic had described Claude as the department’s most widely used frontier model and the only one on classified systems. (congress.gov) The same Congressional Research Service brief said reports tied Claude to the January 2026 operation that captured Venezuelan President Nicolás Maduro, after which Anthropic sought more clarity from the Pentagon about how its models were being deployed. The brief describes the broader fight as a dispute over autonomous weapon systems and model-use restrictions. (congress.gov) Since then, officials across Washington have kept looking for ways to evaluate Mythos. Politico reported on April 14 that staff on at least three congressional committees had held or requested briefings on Mythos’ cyber-scanning capabilities, and that federal officials were still trying to access the model despite the administration’s public posture toward Anthropic. (politico.com) Anthropic’s pitch is easier to explain in cybersecurity terms than in abstract debates about “frontier AI.” A model that can read and modify complex code can act like an automated security reviewer, but the same capability can also help identify offensive paths into critical systems, which is why Anthropic paired the launch with risk documents and a limited rollout. (anthropic.com 1) (anthropic.com 2) The company has not publicly resolved the underlying argument with the Pentagon. But Mythos Preview has given Anthropic a concrete national-security product to put in front of White House officials, lawmakers and agencies that want help securing software they already rely on. (theverge.com) (anthropic.com)