OpenAI signs AI deployment deal with U.S. Department of War, sets strict limits

The company said the arrangement includes more guardrails than previous classified AI deployments and asked the government to offer similar terms to other AI developers.

According to OpenAI, the agreement is guided by 3 firm limits. Its technology cannot be used for mass domestic surveillance, cannot direct autonomous weapons systems, and cannot make high stakes automated decisions such as social credit style evaluations.

The systems will operate only through cloud-based infrastructure controlled by OpenAI, allowing the company to monitor and update safety mechanisms. The firm said it will not provide models without safety protections and will not deploy them on local devices that could potentially be used in lethal autonomous weapons.

The contract also states that the AI system may be used only for lawful purposes and must comply with U.S. laws governing intelligence and surveillance activities. Any handling of private information must follow constitutional protections and existing national security legislation. The system also cannot be used for domestic law enforcement except where permitted by law.

OpenAI said it will place cleared engineers and safety researchers inside government teams to help oversee how the technology is used and to maintain its safety systems.

The company said it decided to proceed with the agreement because it believes the U.S. military needs advanced AI capabilities as rival powers increasingly integrate the technology into their own systems. At the same time, it said the safeguards were designed to ensure that key ethical limits are not crossed.

Earlier, Qazinform News Agency reported that OpenAI outlined steps against misuse of AI tools.