The Pentagon is planning for AI companies to train on classified data, defense official says
- U.S. Department of Defense plans to host classified data in accredited secure facilities for AI model training (per defense official).
- AI companies (OpenAI, Anthropic, xAI) could have limited, cleared personnel access classified data to fine‑tune their models (per official & MIT Technology Review).
- Current use: models like Anthropic’s Claude already answer classified queries, but don’t learn from the data (source: MIT Technology Review).
- Evaluation step: Pentagon will first test model gains using non‑classified data such as commercial satellite imagery before moving to classified sets (per official).
The Pentagon is setting up secure, accredited data centers where AI firms like OpenAI, Anthropic, and xAI can train generative models on classified information, aiming to boost military decision‑making while confronting unprecedented security risks.
