Introduction

The U.S. Department of Defense has quietly integrated frontier AI systems from some of the world's most powerful tech companies into its classified networks, according to newly disclosed contract documents.

What Happened

Through a Freedom of Information Act lawsuit, The Intercept obtained over 400 pages of contract paperwork showing the Pentagon signed deals worth up to $200 million each with OpenAI, Anthropic, Google, and xAI. The agreements call for building AI prototypes designed to improve military advantage, military utility, or enhance military decision making across the armed forces. The contracts include provisions for two-way data sharing, briefings on corporate AI performance, and tabletop exercises simulating real-world combat scenarios.

Anthropic's relationship with the Pentagon soured when the company refused to sign a follow-up amendment banning domestic spying and autonomous weapon use, prompting Secretary of Defense Pete Hegseth to label the firm a supply chain risk. OpenAI, Google, and xAI subsequently moved forward with amended agreements, including embedding company engineers within military commands.

Why This Matters

The documents reveal a level of collaboration between Big Tech and the military that had not been fully public, raising concerns about accountability, ethical boundaries, and the scope of AI use in surveillance and targeting. Experts warn that allowing companies to self-assess risks creates a conflict of interest, especially when the technology is intended for life-or-death military applications.

  • AI systems are being tested for 'warfighting,' automated decision-making, logistics, and military intelligence.
  • Contract clauses require companies to brief the Pentagon on frontier AI model performance and provide risk forecasting exercises.
  • Anthropic's exit highlights the tension between corporate AI ethics pledges and defense demands.

Key Takeaways

Four major AI companies have contractual ties to the Pentagon worth up to $200 million each, with obligations to deliver prototypes, share data, and brief military officials on model performance. Anthropic's departure underscores the growing friction between corporate AI ethics and defense demands. The documents show that AI tools are already being embedded into classified military infrastructure, and oversight mechanisms remain limited and opaque.

Conclusion

As AI capabilities advance, the scope and speed of military adoption will only intensify. The newly revealed contracts offer a rare glimpse into the hidden machinery of defense-tech partnerships, and they underscore the need for greater transparency, independent risk assessment, and public debate over how autonomous systems are authorized to make decisions in conflict zones.