Anthropic's Mythos AI identifies US government security vulnerabilities in testing
Anthropic's Mythos AI system identified numerous vulnerabilities in US government systems during security testing, reportedly breaking into classified systems within hours. The US government responded by blacklisting Anthropic and suspending exports of the Mythos model, citing national security concerns.
What happened
Anthropic conducted testing of its Mythos AI system under Project Glasswing, a limited program to identify and fix vulnerabilities in critical government software before attack. The testing reportedly revealed numerous vulnerabilities in US government systems in a short timeframe.
Senator Mark Warner stated that "Mythos broke into nearly all secret systems, in hours rather than weeks." A government official noted that identifying vulnerabilities did not mean Mythos could simultaneously exploit US security systems.
Reported impact
The vulnerability discoveries raised concerns about the potential for AI systems to compromise sensitive government infrastructure, prompting the Trump administration to blacklist Anthropic and suspend exports of Mythos and Fable models to multiple countries citing national security concerns.
Index analysis
Mythos is designed to detect software vulnerabilities, but cybersecurity experts have warned the technology could also strengthen cyberattacks. The rapid identification of government vulnerabilities demonstrated the system's capability to expose critical security gaps at scale.
Governance controls
The US government ordered Anthropic to halt exports of lighter versions of Mythos due to national security concerns. Anthropic had previously declined to allow the US military to use its AI models for autonomous weapons systems and surveillance operations.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.