Anthropic investigates unauthorized access to Mythos AI system that could enable hacking attacks

Anthropic, the AI safety company behind Claude, discovered a potential breach where someone may have gained unauthorized access to Mythos AI—a system designed to identify security vulnerabilities in software. The company is currently investigating the scope and authenticity of the reported breach.

Mythos AI is built to find exploitable weaknesses in code (the instructions that make software work). If an unauthorized person accessed it, they could use it to identify and exploit security flaws in real-world systems—essentially turning a defensive security tool into an offensive hacking weapon.

For developers and companies relying on Mythos AI for security testing, this raises immediate concerns: if the breach is confirmed, any vulnerabilities they tested through the system may now be known to outsiders, putting their software at risk of attack. Businesses managing sensitive data or critical infrastructure should consider their exposure until Anthropic completes its investigation.