OpenAI Chief Executive Sam Altman informed employees on Thursday that the company is pursuing a potential arrangement intended to help defuse the standoff between Anthropic and the Pentagon over battlefield applications of artificial intelligence.
According to reporting that cited a person familiar with the discussions, no contract has been signed and the negotiations may not lead to an agreement. In a note to staff viewed by the reporting outlet, Altman described work on a possible pact with the Department of War that would allow OpenAI models to be used in classified environments so long as the usage aligns with the companys stated principles.
The draft contract described by Altman would permit government use of OpenAIs models except in instances deemed unlawful or unsuitable for cloud deployment. The memo specifically noted prohibitions such as domestic surveillance and the deployment of autonomous offensive weapon systems without humans in the loop.
Altman conveyed that OpenAI intends to preserve these guardrails primarily through technical measures rather than relying solely on contractual promises. One example cited in the note is a contract provision that would limit operation of the models to cloud-hosted environments, excluding edge deployments. That approach would, in the companys view, help prevent use cases like fully autonomous weapons that do not include human oversight.
In addition to cloud-only restrictions, Altman wrote that OpenAI would develop technical safeguards and place company personnel alongside government teams to help ensure the systems operate as intended. The company said it would offer comparable services to allied governments if the model proves workable, and suggested that such a framework might serve as a template for other AI research labs seeking to reconcile safety policies with government requests.
Earlier on Thursday, Anthropics CEO Dario Amodei announced that his company had rejected the Department of Wars demand to make its technology available for all lawful uses. Anthropic reiterated its intent to retain the right to block uses such as mass domestic surveillance and autonomous weapons.
The discussions reflect an effort by one leading AI developer to position technical controls, cloud-hosting limitations, and operational oversight as mechanisms to permit classified government deployments while limiting applications the company deems unacceptable. Whether these proposals will produce a binding agreement remains uncertain.