Deployment inside NSA
Anthropic has embedded roughly half a dozen engineers at the U.S. National Security Agency to assist with the deployment and customization of its Mythos AI model for offensive cyber operations, according to Financial Times, citing two people familiar with the arrangement. The engineers are described as forward-deployed personnel positioned to guide use of the model and adapt it to particular operational needs.
Legal dispute with Pentagon
The move comes amid an active legal battle between Anthropic and the Pentagon over limitations placed on how the company's Claude models may be employed in warfighting. Anthropic previously sought restrictions intended to prevent government use of Claude for mass surveillance of American citizens and for lethal autonomous drones. In response, the Pentagon designated the AI lab a "supply-chain risk," a classification that Anthropic has challenged in court. That designation carries implications that could compel the company to end contracts with organizations that work with the U.S. military.
Mythos rollout and access
Anthropic introduced Claude Mythos in April, initially making access available only to a limited set of U.S.-based organizations. Earlier this week, the company said it would expand distribution of Mythos to 150 organizations across 15 countries. The announcement followed concerns from various institutions about the model's capabilities.
Concerns and capabilities
Governments, financial institutions and IT companies have flagged Mythos for its potential to detect and exploit software vulnerabilities. One person familiar with the situation told observers that Mythos could be useful for infiltrating the networks of nation-states such as China or Iran. Separately, a person close to Anthropic stated that adversaries are likely working on their own AI-driven offensive tools.
Market and competitive context
Anthropic has filed for an initial public offering that could value the company at more than $1 trillion. Since the release of Mythos, rival OpenAI has introduced a model described as having similar capabilities.
Implications
The presence of Anthropic engineers within a U.S. intelligence agency, ongoing litigation over Pentagon restrictions, broadening access to Mythos, and rapid competitive responses together outline a complex set of technological, legal and market developments. Stakeholders across national security, cybersecurity, financial services and the IT sector are monitoring how deployment practices and regulatory responses evolve.