Pentagon Plans Secure AI Training on Classified Data

The Pentagon is exploring plans to create secure, isolated environments where generative AI companies could train specialized versions of their models on classified military data. This initiative would allow models like Anthropic's Claude to learn from sensitive intelligence, operational reports, and tactical information, thereby enhancing their utility for defense-specific tasks. By providing access within a controlled, air-gapped setting, the military aims to develop AI assistants with deep domain knowledge of military protocols and scenarios without compromising national security. The discussions highlight the growing recognition of AI's strategic importance and the unique infrastructure required to adapt commercial AI breakthroughs for high-stakes national security applications.