Back
Anthropic deploys unreleased Mythos AI with NSA for cyber operations

Anthropic deploys unreleased Mythos AI with NSA for cyber operations

Semafor reported on June 5 that the NSA is using Anthropic’s withheld Mythos model for cyber work, including defense-related uses, as Anthropic warns of rising AI-enabled threats.

Fact Check
Anthropic's official Project Glasswing page confirms Mythos is an unreleased frontier model with cyber capabilities and acknowledges discussions with US government officials. An FT scoop by Demetri Sevastopulo and Cristina Criddle reports NSA is using Mythos for offensive cyber operations with Anthropic engineers embedded at the agency, corroborated by secondary reporting from Crypto Briefing. The claim's attribution to Semafor differs from the FT scoop, but the substantive facts—NSA use of the withheld Mythos model for cyber work, and Anthropic warnings about AI-enabled threats—are supported.
    Reference: 123
Summary

Anthropic is embedding engineers at the U.S. National Security Agency to deploy its unreleased Mythos AI model for cyber operations, according to a June 5 Semafor report. The model was reportedly withheld from public release because of its strong ability to find and exploit software vulnerabilities, but was provided to selected institutions to help build defenses. The disclosure adds a government deployment layer to Anthropic’s broader cybersecurity push and comes as the company has warned that AI is playing a larger role in increasingly sophisticated and more autonomous cyber threats. It also comes amid a legal dispute between Anthropic and the U.S. government.

Terms & Concepts
  • Mythos AI model: Anthropic’s advanced unreleased AI system reportedly used for sensitive cybersecurity work.
  • Cyber operations: Activities conducted in digital networks for security, intelligence, or offensive and defensive cybersecurity purposes.
  • Software vulnerabilities: Weaknesses in software or code that can be identified for defense or exploited in attacks.