The Pentagon has reached a deal with seven of the largest AI labs to deploy their models on classified military systems. The labs include OpenAI, Anthropic, Google DeepMind, Meta, Microsoft, Amazon, and Scale AI.
Key facts
- 7 AI labs including OpenAI, Anthropic, Google DeepMind, Meta, Microsoft, Amazon, Scale AI
- First formal framework for commercial AI on classified military systems
- OpenAI changed its charter in January 2024 to allow military use
- Pentagon created CDAO in 2023 to accelerate AI adoption
The agreement, first reported by @rohanpaul_ai, marks the first formal framework for using cutting-edge commercial AI on classified military networks. The seven labs represent the dominant players in frontier AI development, collectively commanding over $50 billion in funding.
Key Takeaways
- US military deal with 7 AI labs for classified systems.
- First formal framework for commercial AI on classified networks.
What the Deal Covers
The deal allows each lab's models to be used on classified Pentagon systems for tasks including intelligence analysis, logistics planning, and cyber defense. The exact terms remain undisclosed, but sources indicate the agreement includes data security protocols and human oversight requirements.
This arrangement parallels recent moves by the Department of Defense to accelerate AI adoption, including the 2023 creation of the Chief Digital and Artificial Intelligence Office (CDAO). The Pentagon has been testing large language models on unclassified systems since early 2024.
Why This Matters
The unique angle here is the reversal of AI labs' earlier resistance to military applications. OpenAI changed its charter in January 2024 to allow 'military and defense' use cases, while Anthropic has maintained a policy against weapons development but permits defensive applications. Meta's open-source Llama models have been used by military contractors for months.
[According to @rohanpaul_ai], the deal covers 'classified systems' — a significant expansion beyond the unclassified trials. This raises questions about AI safety protocols and the potential for autonomous weapons development, though the labs have publicly stated their models will not be used for autonomous targeting.
Competitive Implications
The agreement gives the Pentagon preferential access to the latest models before they reach civilian markets. This could create a two-tier system where military applications receive priority compute and safety testing, potentially slowing commercial releases.
Scale AI, the data annotation and evaluation company, is an interesting inclusion — it provides the infrastructure for model evaluation rather than frontier models themselves. This suggests the deal encompasses not just model deployment but also testing and validation pipelines.
What to watch
Watch for the release of specific safety protocols and human oversight requirements — likely within 90 days. Also watch whether Anthropic or OpenAI disclose any red-teaming results specific to classified applications, and whether the deal triggers congressional hearings on AI in military systems.
