Anthropic’s Mythos Model Found Vulnerabilities in Classified US Government Systems, Official Says
Source SecurityWeek
Rated Critical by Threat Radar from the SecurityWeek text
Data as of 24 Jun 2026, 03:39 UTC
What to do
AI-suggested
No specific remediation or patch is applicable as this is a report of vulnerability identification by an AI model, not a disclosed software vulnerability with a fix. The U.S. government has imposed access restrictions on Anthropic's AI models to mitigate national security risks. Organizations should follow official government directives regarding the use and access of such AI tools. Continued collaboration between AI developers and government agencies is recommended to balance innovation with security.
Description
Anthropic's Mythos AI model identified vulnerabilities in classified U.S. government computer systems during a cooperative testing exercise called Project Glasswing. The vulnerabilities were found within hours, but it is not confirmed that the model exploited them in that timeframe. The testing involved collaboration between Anthropic and U.S. intelligence agencies to assess critical software security risks. Despite the findings, no public details confirm exploitation or specific vulnerabilities. The U.S. government has imposed restrictions on Anthropic's AI models due to national security concerns. The situation reflects ongoing tensions between Anthropic and the U.S. administration regarding AI use and security.