Anthropic's Mythos model reportedly powers NSA cyber ops against China and Iran

The US National Security Agency (NSA) is actively using Anthropic's advanced AI model, Mythos, for offensive cyber operations against China and Iran, according to a Financial Times report. Anthropic reportedly embedded six of its own engineers at the agency to provide system adaptation and technical support. This partnership unfolds even as the Pentagon has classified Anthropic as a supply chain risk due to ongoing legal disputes. Despite that, the NSA retained exempted access to Mythos, which the company itself had previously deemed too powerful for hacking tasks. Compounding the controversy, an unauthorized group managed to breach Mythos's source code, raising further security concerns.

Meanwhile, Anthropic has expanded access to Mythos to 150 organizations across 15 countries, signaling a shift toward broader deployment of a model originally restricted for its offensive potential. The developments come as President Trump signed a new executive order on AI safety testing, highlighting the growing tension between national security interests and AI governance. The dual-use nature of Mythos — capable of both defensive and offensive cyber operations — underscores the urgent need for clearer boundaries as AI systems enter military theaters.