OpenAI, the San Francisco-based artificial intelligence company, has agreed to deploy advanced AI systems in classified environments for the US Department of Defense, and is calling on the government to extend the same contractual terms to competing laboratories.
The company said the arrangement includes explicit prohibitions on using its technology for mass domestic surveillance, directing autonomous weapons systems, or making high-stakes automated decisions such as social credit scoring.
"We think our agreement has more guardrails than any previous agreement for classified AI deployments, including Anthropic's," OpenAI said, in a direct reference to its rival.
Anthropic, the AI safety company backed by Amazon, has previously disclosed its own national-security partnerships with US government agencies.
Under the terms OpenAI described, deployments will be cloud-only and managed through a safety stack that remains under the company's control, with access limited to security-cleared OpenAI engineers and safety researchers.
Contract language published alongside the announcement restricts use to lawful purposes and requires human oversight wherever law, regulation or policy demands it.
Intelligence activities conducted under the agreement must comply with existing surveillance and privacy authorities, and OpenAI said unconstrained monitoring of US citizens is not permitted under current law.
Related reading
- OpenAI and Amazon announce strategic partnership
- OpenAI and Microsoft confirm they're still partners, as Amazon's new $50bn investment in the ChatGPT owner br…
- OpenAI announces $110bn funding round backed by Amazon and NVIDIA
The company said it retains full discretion over its safety systems and can terminate the contract if the terms are violated.
OpenAI framed the public disclosure as an attempt to set a higher baseline for the industry, urging the US government to make equivalent contractual protections available to other AI developers and to deepen collaboration between the public and private sectors on AI governance.
The recap
- OpenAI agreed classified AI deployment with the Department of War.
- Deployment will be cloud-only and run by OpenAI safety stack.
- OpenAI requested the government make terms available to all labs.
