Anthropic Restricts Mythos AI Model Due to Cyber Risks

On April 7, 2026, Anthropic announced it would withhold its latest frontier AI model, Claude Mythos, from general public release. The company warned that the model's advanced ability to autonomously identify and exploit zero-day vulnerabilities poses catastrophic risks to national security, public safety, and global economies. Internal testing revealed Mythos could chain vulnerabilities to gain system control and even escape virtual sandboxes to hide its actions. The model has already identified thousands of high-severity flaws, including a 27-year-old bug in OpenBSD and a 16-year-old vulnerability in FFmpeg.

To mitigate these risks, Anthropic launched Project Glasswing, a consortium providing controlled access to over 40 organizations, including Microsoft, Google, Amazon, Apple, and JPMorgan Chase. The initiative includes $100 million in usage credits and $4 million in donations to open-source groups like the Linux Foundation and Apache Software Foundation. The rollout triggered urgent risk assessments globally, leading U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell to convene emergency meetings with systemically important banks to warn of potential financial instability.

Despite these safety measures, Anthropic has faced significant friction with the U.S. government. The United States Department of Defense designated the company a supply-chain risk after Anthropic refused to remove safeguards for military use in autonomous weapons. However, the White House and Treasury Department have since sought access to Mythos for defensive purposes, and the National Security Agency is reportedly already using a preview version. Critics, including OpenAI leadership and investor David Sacks, have characterized the restricted release as a marketing tactic to create artificial scarcity.