OpenAI shares more details about its agreement with the Pentagon

In a recent admission, OpenAI's CEO Sam Altman described the company's new agreement with the Pentagon as "definitely rushed" and acknowledged that the situation surrounding the deal lacked positive optics. The deal follows a breakdown in negotiations between Anthropic and the Department of Defense, after which President Donald Trump ordered federal agencies to cease utilizing Anthropic’s technology, labeling it a supply-chain risk. In a surprising turn, OpenAI swiftly announced its own agreement to deploy AI models in classified settings. Both OpenAI and Anthropic have established strict boundaries regarding their technology’s use, particularly in relation to fully autonomous weapons and mass domestic surveillance. This raises pressing questions about the integrity of OpenAI’s stated safeguards and the reasons behind its successful negotiations while Anthropic's efforts failed. As OpenAI executives defended the agreement on social media, the company released a blog post clarifying its stance. The post emphasized three critical areas where OpenAI’s models are not permitted: mass domestic surveillance, autonomous weapon systems, and high-stakes automated decision-making, such as social credit systems. OpenAI asserted that, unlike some competitors who have diminished safety measures, its agreement adopts a comprehensive, layered approach to safeguarding its technologies. The blog stated, "We retain full discretion over our safety stack, we deploy via cloud, cleared OpenAI personnel are involved, and we have robust contractual protections," highlighting adherence to U.S. legal standards. Despite these assurances, some critics, including Techdirt’s Mike Masnick, have suggested that the agreement could still facilitate domestic surveillance, pointing to compliance with Executive Order 12333, which has been associated with NSA surveillance tactics. In response to this criticism, Katrina Mulligan, OpenAI’s head of national security partnerships, emphasized that discussions about the contract language overlook the importance of deployment architecture. She argued that by confining their deployment to cloud API, OpenAI can prevent integration into weapons systems or operational hardware. Altman further addressed concerns about the rushed nature of the deal, which has sparked considerable backlash against OpenAI, even leading to Anthropic's Claude surpassing ChatGPT in Apple’s App Store. When questioned about the rationale behind the agreement, Altman stated, “We wanted to de-escalate tensions, believing that the deal could contribute positively to the relationship between the Department of War and the tech industry. If successful, we’ll be seen as pioneers; if not, we may face ongoing criticism for being hasty and careless.”