Altman: Pentagon Deal Looked 'Sloppy'

The agreement to deploy OpenAI's models on the Pentagon's classified networks was announced on February 27, 2026, just hours after the Pentagon ended its collaboration with AI competitor Anthropic. Anthropic had reportedly refused to budge on its "red lines" concerning the use of its AI in surveillance and autonomous weapons. In January 2024, OpenAI quietly updated its usage policy, removing a blanket ban on "military and warfare" applications. The previous policy explicitly forbade "activity that has high risk of physical harm," including "weapons development" and "military and warfare." The revised policy has a more general prohibition against using its services to "harm yourself or others," which includes developing or using weapons. The initial deal with the Pentagon faced immediate backlash over concerns that it lacked explicit prohibitions on domestic mass surveillance and the development of autonomous weapons. Critics pointed out that relying on existing U.S. law was insufficient, as those same legal frameworks had been used to justify broad surveillance programs in the past. In response to the criticism, OpenAI and the Pentagon amended the agreement to include more specific language. The new clauses explicitly forbid the use of the AI for intentional domestic surveillance of U.S. persons and clarify that intelligence agencies like the NSA cannot use OpenAI's systems without a separate agreement. The fallout from the announcement was significant, with reports of a 295% surge in ChatGPT uninstalls over the subsequent weekend. Simultaneously, Anthropic's competing chatbot, Claude, saw a significant increase in downloads and became the number one app on the U.S. App Store. Internally, the deal also caused friction. Aiden McLaughlin, an OpenAI scientist, publicly slammed the agreement as "not worth it." The criticism from both inside and outside the company highlighted the ethical tightrope OpenAI is walking as it pursues lucrative government contracts. Sam Altman, in his candid remarks, stated that rushing the announcement was an attempt to "de-escalate" a tense situation between the Pentagon and the broader AI industry. He expressed concern that the path things were on was "dangerous for Anthropic, healthy competition and the U.S." and that OpenAI had negotiated for similar terms to be offered to other AI labs.