OpenAI is negotiating additional safeguards with the US defense department to prevent mass surveillance of American citizens using its AI technology, following a hastily announced deal last Friday. The company has already modified contractual language around surveillance and is working to add protections during the three-month implementation period.

Legal experts and staff have scrutinized language in the contract that prohibits "intentional," "deliberate," or "targeted" surveillance, raising concerns that the government could surveil Americans "incidentally" or "unintentionally" using modern AI tools. OpenAI's approach contrasts with rival Anthropic, which refused contract terms due to surveillance concerns. The effort to add protections comes as OpenAI seeks to clarify terms and address staff concerns about potential abuse of its powerful AI.

OpenAI has argued it could maintain redlines on surveillance through technical measures and by ensuring employees remain "in the loop" with officials. However, Anthropic CEO Dario Amodei dismissed these protections as ineffective, writing that OpenAI "cared about placating employees" while his company "actually cared about preventing abuses." The negotiations highlight the tension between national security applications and civil liberties protections as AI technology advances rapidly.