The Pentagon is making plans for AI companies to train on classified data, defense official says
Defense officials aim to embed sensitive battlefield intel directly into models like Claude for more accurate targeting.
The Pentagon is actively discussing plans to establish secure, classified environments where leading generative AI companies can train specialized military versions of their models. According to a defense official speaking to MIT Technology Review, this would allow firms like Anthropic (Claude), OpenAI, and Elon Musk's xAI to train their AI directly on top-secret data, including surveillance reports and battlefield assessments. The goal is to create models that are far more accurate and effective for specific defense tasks, such as analyzing targets in Iran, moving beyond the current use of AI models that merely answer questions in classified settings.
This initiative represents a significant shift, as it would embed sensitive intelligence directly into the models' training data, presenting unique security challenges. It also brings AI firms into much closer contact with classified information than ever before. The news comes as the Department of Defense implements a new agenda to become an 'AI-first' warfighting force and has already secured agreements with OpenAI and xAI to operate their models in classified environments. The push for more powerful, domain-specific military AI is accelerating amid escalating conflict with Iran.
- The Pentagon plans secure facilities for AI firms to train models on classified military intelligence.
- Models like Anthropic's Claude would learn from sensitive data like surveillance reports to improve target analysis accuracy.
- The move is part of a broader 'AI-first' warfighting strategy as tensions with Iran escalate.
Why It Matters
This could create a new class of military-specific AI but raises profound security risks about embedding state secrets into commercial models.