NSA using Anthropic's Mythos despite blacklist

A new report from Axios reveals that the National Security Agency (NSA) is actively using Anthropic's advanced Mythos AI model for classified intelligence work, directly contravening a Pentagon-wide blacklist. The policy, established over security concerns about commercial large language models (LLMs), explicitly prohibits Department of Defense components from using models like Anthropic's. Despite this, the NSA, which operates under the Department of Defense but also serves as a national intelligence agency, has proceeded with deploying Mythos for its superior reasoning and analysis capabilities on sensitive data.

The internal conflict underscores the tension between stringent security protocols and the operational demand for cutting-edge AI tools. The NSA's move suggests that the perceived intelligence advantages of models like Mythos—likely used for tasks such as summarizing foreign communications, detecting patterns in vast datasets, or generating analytical reports—outweigh the mandated security risks associated with commercial AI. This situation points to a potential loophole or differing risk assessment within the intelligence community compared to the broader military establishment, and could force a reevaluation of the blanket ban as AI becomes increasingly critical to national security missions.