OpenAI Publishes Pentagon Contract 'Red Lines'
Following its deal with the Pentagon, OpenAI has publicly released the contract language, a move aimed at transparency. The terms include explicit "red lines" prohibiting use for mass domestic surveillance or autonomous weapons, and require human-in-the-loop safeguards.
This deal materialized after rival AI lab Anthropic refused a similar contract, leading the Trump administration to order federal agencies to cease using Anthropic's technology and declare it a "supply chain risk". OpenAI announced its agreement shortly after, stating a desire to "de-escalate" the situation between the U.S. government and AI labs. The core of the disagreement for Anthropic reportedly centered on the Pentagon's insistence on an "all lawful purposes" clause, which Anthropic argued could contain legal loopholes. OpenAI accepted this language, expressing comfort with relying on existing laws, a point of differentiation from Anthropic which sought more explicit prohibitions written directly into the contract. OpenAI asserts its agreement has more robust safeguards than what was offered to Anthropic, citing a multi-layered approach. This includes a cloud-only deployment architecture, which OpenAI says prevents use for fully autonomous weapons that would require edge deployment, and the retention of full control over its safety stack. The contract also stipulates that cleared OpenAI personnel will remain in the loop on the use of the technology within classified environments. Sam Altman has stated that if the government were to violate the terms, OpenAI could terminate the contract, though he does not expect that to happen. The deal has sparked significant debate, with some critics and even OpenAI employees raising concerns over the ethics of military AI. In response to the backlash, Anthropic's Claude app saw a surge in popularity, reaching the number one spot on the App Store ahead of ChatGPT. OpenAI had previously partnered with defense tech startup Anduril Industries to use its models for defending against drones. The new Pentagon agreement is part of a broader trend of major tech companies, including Google and Palantir, increasingly working with the U.S. military as the financial costs of training large-scale AI models rise. As part of the agreement, OpenAI urged the government to resolve its issues with Anthropic and not designate it as a "supply chain risk," a label typically reserved for foreign adversaries. Sam Altman also noted that the terms offered to OpenAI would be made available to all other AI labs.
