Anthropic's Mythos identified vulnerabilities in NSA systems during testing
Anthropic's Mythos model identified vulnerabilities in US government classified computer systems during a security testing programme, reportedly breaking into systems within hours. The US government subsequently ordered Anthropic to suspend exports of its advanced models worldwide over national security concerns.
What happened
Anthropic's Mythos model identified vulnerabilities in highly sensitive US government computer systems during a testing exercise called Project Glasswing, which was designed to find and fix vulnerabilities in critical software before attackers could exploit them. Senator Mark Warner reported that NSA chief Joshua Rudd stated Mythos 'broke into almost all of our classified systems, not in weeks, but in hours.'
According to an unidentified US official cited by the Associated Press, although Mythos identified certain vulnerabilities within hours, that did not mean the model was able to exploit them within that time.
Reported impact
The White House pulled Mythos 5 and Fable 5, Anthropic's two most advanced models, from public access over national security concerns about what they could achieve in foreign hands.
The US government ordered Anthropic to suspend exports of its latest Mythos and Fable models to destinations worldwide and all foreign nationals, citing national security concerns.
Index analysis
Frontier AI models are described as poised to transform offensive and defensive cyber capabilities on a timeline measured in months, according to a rare joint warning from the cyber agencies of the Five Eyes alliance.
Governance controls
Project Glasswing was a restricted government program designed in partnership with Anthropic and intelligence agencies to test for vulnerabilities in critical systems before attackers could exploit them.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.