OpenAI makes changes to 'opportunistic and sloppy' Pentagon deal - FT

OpenAI makes changes to 'opportunistic and sloppy' Pentagon deal - FT

OpenAI Revises Pentagon Agreement Amid Scrutiny Over Safeguards

OpenAI has finalized a revised agreement with the U.S. Department of War (DoW) to deploy its advanced AI systems in classified environments, addressing concerns over ethical and legal boundaries. The deal, announced February 28, includes explicit safeguards against uses such as autonomous weapons, mass domestic surveillance, and high-stakes automated decision-making, aligning with red lines previously outlined by the company.

The agreement emphasizes a cloud-only deployment model, retaining OpenAI's proprietary safety stack and requiring cleared personnel to oversee implementation. Contractual language prohibits the AI system from independently directing autonomous weapons or handling domestic law enforcement activities without compliance with the Posse Comitatus Act. Additionally, the company asserts that existing U.S. laws, including the Fourth Amendment and DoD Directive 3000.09, further restrict the DoW's use of the technology.

Critics, however, argue the deal contains loopholes. Jessica Tillipman, a government procurement law expert, noted that the contract does not grant OpenAI an "Anthropic-style, free-standing right to prohibit otherwise-lawful government use" according to analysis. This has raised concerns that the agreement relies heavily on the assumption that the DoW will adhere to current laws, despite historical precedents like the Snowden revelations, where legal surveillance practices were later deemed unlawful as reported.

The deal also follows a contentious standoff between the Pentagon and Anthropic, which refused to remove safety guardrails for its AI model Claude. Defense Secretary Pete Hegseth condemned Anthropic's stance as "arrogance and betrayal," prompting threats to classify the company. OpenAI's approach, by contrast, emphasizes legal compliance and collaboration, with CEO Sam Altman stating the company sought to "de-escalate tensions" while ensuring "responsible safeguards" according to company statements.

OpenAI's revised terms aim to balance national security needs with ethical constraints, but questions remain about enforceability in classified settings. The company has not disclosed how its safety protocols differ from standard user applications, nor how it will verify compliance in real-time as noted in analysis. As the Pentagon accelerates its AI integration, the outcome of this agreement could set a precedent for future partnerships between AI firms and government agencies.

(https://openai.com/index/our-agreement-with-the-department-of-war/): OpenAI's blog post outlines contractual terms and safety measures.
(https://www.technologyreview.com/2026/03/02/1133850/openais-compromise-with-the-pentagon-is-what-anthropic-feared/): Analysis from Technology Review highlights legal and ethical critiques.