Mythos finds vulnerabilities in classified US systems during testing
An official confirms vulnerability discovery; Warner’s broader breach claim receives subsequent clarification.
AnthropicA US official told AP that Mythos identified vulnerabilities in classified government systems during joint testing with Anthropic. Discovery within hours did not mean exploitation within hours. The source spoke anonymously; NSA and Anthropic declined comment. Associated Press
Earlier, Senator Mark Warner said NSA and Cyber Command chief Joshua Rudd had told him the model broke into almost all classified systems within hours. IBTimes reported that account on 21 June, while acknowledging the absence of public agency confirmation and details of the affected systems. IBTimes UK
The Economist reporter who carried Warner’s remarks, Shashank Joshi, subsequently warned against reading the wording literally. He said the result would depend on other tools and particular conditions, and acknowledged that his initial report lacked caveats. Straight Arrow News reported that clarification. Straight Arrow News
Taken together, these accounts support a reported testing result while leaving the extent of successful exploitation unresolved. They do not establish the sweeping autonomous breach suggested by the headline. Associated Press Straight Arrow News
Sources & attribution
- Independent investigation 24 Jun 2026Anthropic’s Mythos model found vulnerabilities in classified US government systems, official says
Associated Press. Reporting based on an anonymous US official. Publication date in UTC.
- Analysis 21 Jun 2026Anthropic’s Mythos AI Cracked Almost All US Classified Systems in Hours, NSA Chief Reportedly Said
IBTimes UK. Warner’s reported account; no public agency confirmation.
- Analysis 22 Jun 2026No, the NSA wasn’t hacked by AI. Here’s what actually happened
Straight Arrow News. Reports Shashank Joshi’s clarification of his original wording.