Pentagon plans to weaponize Anthropic's Claude Mythos Preview in new cyber task force

A leaked email and two anonymous sources cited by Politico reveal that the Pentagon is actively planning to adopt and weaponize frontier AI models with hacking capabilities, specifically Anthropic's unreleased Claude Mythos Preview. This initiative comes despite the Pentagon officially designating Anthropic a supply chain risk after a public disagreement, theoretically barring collaboration with the company. The task force, announced by Joshua Rudd—who heads both U.S. Cyber Command and the NSA—aims to study how leading AI models can be deployed safely across military missions, including on high-side systems handling the intelligence community's most sensitive secrets.

While part of the effort echoes Anthropic's Project Glasswing (which allows approved organizations to use Claude Mythos Preview for vulnerability detection and patching), the task force also has an offensive dimension. Former Cyber Command deputy commander Charles Moore stated that AI tools are essential for both defensive and offensive cyber operations. Despite its feud with Anthropic, the White House reportedly retained veto power over expanding Project Glasswing to 70 new companies. Anthropic has previously indicated openness to using Mythos in offensive operations. Politico notes the Pentagon is likely to explore weaponizing the model.