OpenAI details the multi-layered protection measures in its agreement with the U.S. Department of Defense.

OpenAI said on Saturday that the agreement it reached with the U.S. Pentagon the previous day to deploy technology on the Department of Defense’s classified networks includes additional safeguards to standardize technology usage scenarios. U.S. President Trump ordered the government to stop cooperation with Anthropic on Friday, and the Pentagon announced that the startup was listed as a supply chain risk. The move is a major blow to the artificial intelligence lab following a showdown over technical security measures. Anthropic said it would litigate any risk determination in court.

Soon after, OpenAI, a competitor backed by investors such as Microsoft, Amazon, and SoftBank, also announced its own cooperation agreement on Friday night.

OpenAI said on Saturday: "We believe that our agreement sets more security protection constraints than any previous confidential artificial intelligence deployment agreement, including Anthropic."

The artificial intelligence company said that the contract signed with the U.S. Department of Defense, which has been renamed the "Department of War" by the Trump administration, sets three red lines:

OpenAI technology must not be used for domestic mass surveillance in the United States, command autonomous weapons systems, or any high-risk automated decision-making.

OpenAI said: "In the agreement, we protect these red lines through a more comprehensive and multi-layered approach. We have full decision-making power over our own security system, the technology is deployed through the cloud, security-vetted OpenAI personnel participate throughout the process, and we have strict contractual protection clauses."

In the past year, the Pentagon signed individual agreements with a maximum value of US$200 million with major artificial intelligence laboratories such as Anthropic, OpenAI, and Google. The Pentagon wants to retain full flexibility in defense, unencumbered by warnings from technology developers that unreliable artificial intelligence should not be used in weapons.

OpenAI warned that if the U.S. government breached the terms of the contract, it could lead to the termination of the agreement, but added: "We do not expect this to happen."

The company also said its rival Anthropic should not be labeled a "supply chain risk," noting: "We have made our position clear to the government."