Anthropic’s Mythos AI Reportedly Hacked the NSA’s Most Sensitive Systems ‘in Hours’

Key Points:

• Anthropic's AI model Mythos, designed to find and exploit cybersecurity vulnerabilities, caused alarm in the sector, leading the company to restrict its release to select testers including the NSA.
• Reports emerged that the NSA found multiple vulnerabilities in its own systems during tests with Mythos, sparking fears about global cybersecurity defenses, though officials clarified these tests were conducted in highly controlled environments and vulnerabilities were identified but not exploited.
• Senator Mark Warner claimed Mythos breached almost all NSA classified systems within hours, citing information from NSA head General Joshua Rudd, which intensified public concern and prompted a multinational intelligence coalition to call for a comprehensive societal response to AI cybersecurity risks.
• The Trump administration restricted access to Mythos-class models for foreign nationals citing national security, a move criticized for potentially weakening U.S. cybersecurity and benefiting adversaries, and reportedly halted NSA's internal Mythos testing.
• Following the ban, the administration is reportedly negotiating with Anthropic to restore limited NSA access for national security purposes, while clarifications suggest the initial fears about Mythos’ impact may have been exaggerated.