OpenAI's Pentagon Deals Draw Scrutiny
OpenAI CEO Sam Altman admitted the company's recent Pentagon deal “looked opportunistic and sloppy.” The controversy comes as OpenAI also struck an agreement with the DoD's Office of Warfighting to deploy AI models on classified networks, deepening its role in national security.
OpenAI's work with the U.S. military follows a significant policy shift in January 2024, when the company removed language from its usage policy that explicitly banned "military and warfare" applications. This change replaced a clear prohibition with a vaguer rule against using its services to cause harm, paving the way for defense-related contracts. The controversy intensified after OpenAI's rival, Anthropic, refused a Pentagon deal over concerns it would permit mass surveillance and fully autonomous weapons. Anthropic had insisted on writing specific safeguards into its contract, which the Department of Defense declined. Subsequently, the Trump administration ordered federal agencies to cease using Anthropic's products, designating the company a "supply chain risk." Within hours of Anthropic's relationship with the Pentagon imploding, OpenAI announced its own agreement. The initial terms, which allowed for "any lawful applications," drew immediate criticism for being broad enough to potentially include the very uses Anthropic had rejected. Critics noted that existing legal frameworks have been used to justify expansive surveillance programs in the past. In response to the backlash, OpenAI and the Pentagon amended the agreement to explicitly prohibit the intentional domestic surveillance of U.S. persons. The updated contract also specifies that intelligence agencies like the NSA cannot use OpenAI's systems without a separate agreement. Despite these revisions, some critics argue the safeguards still rely on the government's interpretation of what is "lawful." The deal has not been without internal dissent. At least one OpenAI researcher publicly stated the deal was not "worth it," while another called the contractual safeguards "window dressing." The move also prompted a public response, with reports of a significant spike in uninstalls of ChatGPT and an open letter from hundreds of employees at major tech companies, including OpenAI, urging their employers to refuse the Pentagon's demands.
