OpenAI Inks Pentagon Deal with 'Safeguards'
OpenAI has secured a deal to provide its AI systems to the Pentagon, but only after embedding strong "technical safeguards" to prevent misuse. The partnership was announced just hours after the Trump administration banned rival AI firm Anthropic from federal contracts over its refusal to remove ethical restrictions on military use.
The deal follows a quiet but significant change to OpenAI's usage policy in January 2024, when the company removed its explicit ban on "military and warfare" applications. This policy shift replaced a specific prohibition with a broader injunction against using its tools to cause harm or develop weapons, opening the door for non-combat defense partnerships. The central conflict that sidelined Anthropic revolved around the Pentagon's insistence on using AI for "all lawful purposes." Anthropic refused, arguing current laws contain loopholes that could permit mass domestic surveillance through the analysis of commercially available datasets, a use case its CEO Dario Amodei said wasn't illegal, just "never useful before the era of AI." In response to Anthropic's refusal, President Trump directed all federal agencies to cease using the company's technology, setting a six-month phase-out period. Defense Secretary Pete Hegseth then designated Anthropic a "supply chain risk," a label typically reserved for foreign adversaries, effectively blacklisting it from military contracts. OpenAI navigated the impasse by accepting the "all lawful use" clause while negotiating the right to embed its own technical safeguards directly into the AI models deployed on the Pentagon's classified networks. CEO Sam Altman stated these safeguards ensure the models behave as intended and were something the Defense Department also wanted. The agreed-upon red lines mirror the very principles Anthropic advocated for: a prohibition on use for domestic mass surveillance and a guarantee of human responsibility for the use of force, which includes a ban on fully autonomous weapons systems. The safeguards will be deployed on cloud networks only, not on "edge" systems like drones or aircraft. As part of the agreement, OpenAI will embed its own cleared personnel and engineers alongside government teams to help deploy the models and ensure safety protocols are followed. Altman said he urged the Pentagon to offer these same terms to all other AI labs in an effort to "de-escalate" the tensions between the government and the tech industry.
