AP: AI company Anthropic found vulnerabilities in classified US…
AP: AI company Anthropic found vulnerabilities in classified US government systems. A Mythos AI model developed by Anthropic discovered security flaws in highly classified US government systems.
Consensus
- The AI model Mythos developed by Anthropic found vulnerabilities in U.S. government systems
- The discovery was made during a joint testing effort with U.S. intelligence agencies
- The U.S. government restricted access to the Mythos and Fable 5 models for foreign users
- The model was reportedly capable of identifying vulnerabilities quickly, but not exploiting them
Points of divergence
- The testing was part of a project called Project Glasswing involving hundreds of organizations from multiple countries — kommersant
- The model was able to hack into secret systems of the NSA, not just identify vulnerabilities — vm
- There is ongoing debate in the U.S. about restricting the export of advanced AI models — vm
- Kaspersky's founder, Eugen Kaspersky, stated that AI will not be able to take control from humans — vm
Coverage (2 sources)
- AI company Anthropic finds vulnerabilities in classified US government systems — Коммерсантъ
- The Economist: AI model Mythos hacked secret NSA systems in hours — Вечерняя Москва