Anthropic’s Mythos Model Found Vulnerabilities In Classified US Government Systems
Anthropic’s Mythos Model Found Vulnerabilities In Classified US Government Systems
Publish Date: 2026-06-24 02:46:00
Source Domain: www.ndtv.com
- Anthropic’s AI model identified vulnerabilities in highly sensitive government computer systems during a joint testing exercise with US intelligence agencies.
- The testing was conducted under Anthropic’s Project Glasswing initiative aimed at securing critical software to prevent potential fallout from the AI’s capabilities.
- Democratic Sen. Mark Warner mentioned that the AI model, Mythos, penetrated almost all classified systems within hours.
- Tensions persist between Anthropic and the Trump administration concerning AI utilization and security.
- The administration recently required Anthropic to restrict foreign usage of advanced AI models, citing national security concerns.
- Anthropic disabled its customer access to the models in compliance but disagreed with the government’s assessment of the security risks.
- A group of cybersecurity experts argued that the Trump administration’s directive could inadvertently benefit US adversaries, urging a cautious approach.
