Anthropic Mythos reportedly spotted vulnerabilities in highly classified U.S. systems during government-backed testing
apnews.com

Anthropic Mythos reportedly spotted vulnerabilities in highly classified U.S. systems during government-backed testing

Tech News••
2 min read

Published by AINave Editorial • Reviewed by Ramit

TL;DRA U.S. official says Anthropic's Mythos AI model identified vulnerabilities in highly sensitive government computer systems within hours during a testing exercise called Project Glasswing. The test involved collaboration with U.S. intelligence agencies and followed new government directions on vetting advanced AI capabilities.
Anthropic’s Mythos AI model reportedly found vulnerabilities in highly sensitive U.S. government computer systems within hours during a testing exercise. The tests were conducted under an Anthropic initiative called Project Glasswing, which brought together tech firms and government partners to assess how advanced AI could pose risks to public safety, national security, and the economy. The disclosure, referenced in a Senate hearing, comes amid ongoing tensions between AI developers and federal policy on model access and foreign-use restrictions. In response to a government directive, Anthropic disabled its Mythos and related models (Fable 5 and Mythos 5) for all customers. The AP report notes that while vulnerabilities were identified, there is no evidence that they were exploited within the same time frame. The episode has stirred discussion among cybersecurity leaders about the benefits and risks of powerful AI tools for security testing and the need for governance of high-capability models. The story is grounded in briefings and multiple outlets, and details may evolve as investigations continue. For context, the allowed sources referenced include coverage from AP News, which first reported the assertion, and broader coverage documenting the government’s vetting framework for national-security risks in AI. See the FAQs for more on Mythos, Project Glasswing, and related policy questions.

Sources

Latest Tech News