Anthropic's Claude Mythos Finds Vulnerabilities in Classified US Systems
The test shows that a frontier AI can identify security flaws in highly sensitive government networks, raising questions about the pace of AI regulation and the balance between offensive and defensive cyber capabilities.
Reporting from 1 sources: GIGAZINE.
Anthropic's AI Claude Mythos discovered vulnerabilities in classified US government systems during a test called Project Glasswing. The AI found specific flaws within hours, though officials said that does not mean it could have immediately exploited them. The US government has suspended Claude Fable 5 and Claude Mythos 5.
Anthropic's AI Claude Mythos found vulnerabilities in classified US government systems as part of a test initiative called Project Glasswing, according to an anonymous government official who spoke to AP News. The AI identified specific flaws within hours, though the official cautioned that finding a vulnerability does not mean the AI could have immediately exploited it. The US government has ordered the suspension of Claude Fable 5 and Claude Mythos 5. Anthropic has not disclosed what vulnerabilities were found, and relevant agencies have not commented, so specific details remain unknown.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.
Sources
- GIGAZINE Claude Mythosがアメリカ政府の機密システムの脆弱性を発見