OpenAI Amends Pentagon Deal with Safeguards
OpenAI has amended its deal with the Pentagon to include expanded safeguards against its AI being used for mass surveillance or fully autonomous weapons. The move follows internal and public pushback and adds explicit protections to the controversial partnership.
The controversial deal materialized after rival AI company Anthropic refused a Pentagon contract that lacked explicit safeguards against mass surveillance and fully autonomous weapons. After talks collapsed, the Pentagon designated Anthropic a "supply-chain risk" and within hours, OpenAI announced it had secured its own agreement. Initially, OpenAI's agreement permitted the Pentagon to use its technology for "all lawful purposes," which critics noted could include broad surveillance programs already permitted under existing U.S. law. This led to significant public backlash, with "Cancel ChatGPT" trending online and some users switching to Anthropic's Claude, which briefly surpassed ChatGPT in the App Store. The amended contract now explicitly states the AI system "shall not be intentionally used for domestic surveillance of U.S. persons and nationals," referencing the Fourth Amendment and other acts. The new language also expressly prohibits use by intelligence agencies like the NSA under the current agreement, which would require a separate contract. OpenAI CEO Sam Altman acknowledged the rollout of the deal was "rushed" and looked "opportunistic and sloppy," and stated the new terms were intended to make the company's principles "very clear." He also mentioned that OpenAI would work with the DoD on technical protections and would refuse to comply with unconstitutional orders. This incident highlights a growing tension in the AI industry, where major players are navigating lucrative defense contracts while trying to manage ethical concerns and employee dissent. Google previously backed away from Pentagon AI work in 2018 after employee protests, but has since re-engaged in military projects through its cloud division. The updated agreement includes language that any use of AI in autonomous systems must undergo rigorous testing and that the AI cannot be used to "independently direct autonomous weapons" where human control is required by law or policy. OpenAI will also have engineers with security clearances working with the military to monitor the system's use.
