OpenAI Details 'Safeguards' on Pentagon AI

OpenAI's partnership with the Pentagon follows the U.S. government's decision to ban federal agencies from using AI from competitor Anthropic. The ban, which designated Anthropic a "supply chain risk," came after the company reportedly refused to remove restrictions on the use of its AI for domestic mass surveillance and fully autonomous weapons systems. In contrast, OpenAI has established what it calls "red lines" for its military collaboration, which it claims are protected by a multi-layered approach. These include prohibitions on using their AI for mass domestic surveillance, directing autonomous weapons, or making high-stakes automated decisions. To enforce these limitations, OpenAI's AI models will be deployed on a cloud-only basis, preventing use on edge devices that could power autonomous weapons. Additionally, cleared OpenAI personnel will be involved in the deployment and have oversight of the AI's use, and the company states it retains full control over its safety features. The agreement is part of a broader push by the Pentagon to integrate advanced AI capabilities, with the Department of Defense awarding contracts worth up to $200 million each to several major AI firms, including OpenAI, Google, Anthropic (before the ban), and xAI. This initiative aims to use AI for a range of applications from administrative work and data analysis to supporting cybersecurity and warfighting functions. OpenAI's involvement with the military marks a significant shift from its previous policies. In January 2024, the company removed an explicit ban on "military and warfare" applications from its usage policy, replacing it with a more general prohibition against using its services to cause harm. This change paved the way for its current engagement with the Department of Defense. The deal has sparked a debate over who controls the ethical boundaries of military AI – the tech companies developing the models or the government deploying them. While OpenAI asserts its safeguards are robust, some critics point to contract language that allows for "all lawful purposes," which they argue could create loopholes. Defense Secretary Pete Hegseth accused Anthropic of attempting to "strong-arm" the military with its restrictions, a stance he deemed "fundamentally incompatible with American principles." In response to its ban, Anthropic has vowed to challenge the "supply chain risk" designation in court, calling the move "legally unsound." OpenAI has publicly stated that it does not believe Anthropic should be designated a "supply chain risk" and has urged the government to de-escalate the situation. The company has also said it requested the Pentagon make the same terms offered to OpenAI available to all AI companies.