Anthropic Mythos AI finds vulnerabilities in govt systems

An advanced artificial intelligence model developed by Anthropic identified vulnerabilities in highly sensitive U.S. government computer systems within hours during an authorized security test, according to officials. The testing, conducted under a program called Project Glasswing, involved collaboration between the AI company and U.S. intelligence agencies to evaluate the security of classified systems.

The capabilities of the model, named Mythos, were highlighted publicly by Senator Mark Warner of Virginia during a Senate committee hearing on June 11. Warner stated that the head of the National Security Agency and U.S. Cyber Command, General Joshua Rudd, had informed him the tool "broke into almost all of our classified systems, not in weeks, but in hours." Officials clarified that while the model identified certain vulnerabilities rapidly, this did not mean it was able to actively exploit them within that timeframe. The exercise was described as a controlled, internal red-teaming effort designed to find and fix software flaws before malicious actors could.

The revelation comes amid escalating tensions between Anthropic and the federal government. Earlier this month, the...