Anthropic AI model found vulnerabilities in U.S. government systems

During a test with intelligence agencies, Anthropic’s Mythos model found vulnerabilities in highly sensitive U.S. government computer systems, and a U.S. official said it identified certain weaknesses within hours. The official drew a sharp line between finding flaws and exploiting them, saying fast detection did not necessarily mean the system could be broken into that quickly.

Project Glasswing launched on April 7, 2026, to harden critical software against harm to public safety, national security and the economy. Launch partners included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA and Palo Alto Networks. Anthropic later extended access to more than 40 additional organizations and committed up to $100 million in usage credits plus $4 million in donations to open-source security groups.

By May 22, about 50 partners had used Claude Mythos Preview to identify more than 10,000 high- or critical-severity vulnerabilities. The findings covered every major operating system and every major web browser.

Anthropic’s research found Mythos Preview could, when directed by a user, identify and exploit zero-day vulnerabilities in major operating systems and browsers. It also found some of the bugs dated back 10 to 20 years, including a now-patched 27-year-old OpenBSD flaw, the oldest vulnerability uncovered so far.

Sen. Mark Warner introduced the Combat Emerging Threats to Critical Infrastructure Act of 2026 on June 10, saying some critical-infrastructure cybersecurity plans have gone more than a decade without updates even though National Security Memorandum 22 calls for reviews every two years. Warner’s bill would push the Cybersecurity and Infrastructure Security Agency to account for AI-enhanced cyberattacks, AI supply-chain vulnerabilities, deepfakes, robotics and quantum threats.

The National Security Agency declined to comment, and Anthropic declined to discuss the details publicly.

A March 11 designation labeled Anthropic a supply-chain risk after the company refused to remove guardrails from its models for military use, and Anthropic sued the Defense Department and other agencies after that move.