Pentagon pushes AI companies to deploy unrestricted models on classified military networks

Earlier this week, OpenAI signed an agreement for the unclassified network genai.mil, which serves more than three million Department of Defense employees. Many of the usual usage restrictions were lifted as part of the deal, though some safeguards remain in place. Google and xAI have struck similar agreements. According to OpenAI, expanding to classified networks would require a separate agreement.

Anthropic resists dropping safety restrictions

Negotiations with Anthropic are proving significantly more difficult. The company is the only one whose AI chatbot Claude is already available on classified networks through third-party providers, but it refuses to allow its technology to be used for autonomous weapons control and domestic surveillance. At the same time, Anthropic has said it wants to help the US maintain its lead in AI.

AI researchers warn about the risks involved: AI chatbots still hallucinate, and in sensitive military environments, those errors could have deadly consequences. AI companies try to limit risks through built-in safeguards and usage policies. The Pentagon sees it differently - according to Reuters, military officials are frustrated by these company-imposed restrictions. From their perspective, the military should be free to use commercial AI tools however it sees fit, as long as the use complies with US law. Additional rules set by the companies themselves, they argue, are unnecessary.