OpenAI Inks Pentagon Deal as Anthropic Blacklisted
In a major shake-up for defense AI, OpenAI has secured a Pentagon contract after rival Anthropic was designated a "supply chain risk." The move came after Anthropic reportedly refused to give the DoD full control over its models, including source code access. OpenAI CEO Sam Altman publicly committed to building in "technical safeguards" to align with DoD's override and responsible AI requirements, effectively replacing Anthropic as a key partner.
The "supply chain risk" designation applied to Anthropic is a measure historically reserved for foreign adversaries like China's Huawei and Russia's Kaspersky Lab, making its application to a U.S. company unprecedented. This move could bar tens of thousands of government contractors from using Anthropic's AI, potentially posing an existential threat to its public sector business ahead of an anticipated IPO. The Pentagon's demand for unrestricted access to Anthropic's models, including for mass surveillance and autonomous weapons, was the central point of contention. While the DoD publicly denied these intentions, Anthropic held that acceding to such demands would violate its core principles, leading to the impasse and subsequent blacklisting. This contract is part of a broader push by the Pentagon to accelerate AI adoption, managed by the Chief Digital and Artificial Intelligence Office (CDAO). The CDAO is spearheading initiatives to integrate commercial AI across the department, from enterprise business processes to warfighting applications, with recent contracts also awarded to Google and xAI. The agreement with OpenAI falls under the DoD's Responsible AI framework, which mandates that systems be responsible, equitable, traceable, reliable, and governable. OpenAI's public commitment to "technical safeguards" directly addresses these principles, particularly the requirement for human judgment and the ability to disengage systems demonstrating unintended behavior. This deal highlights the critical role of programs like the Small Business Innovation Research (SBIR) program as a pathway for tech companies to engage with the DoD. The Army and Air Force have recently used SBIR solicitations to seek novel AI/ML solutions for applications ranging from decision support to personnel retention, signaling a key entry point for startups in the defense sector. The effort to field AI capabilities connects to long-standing DoD initiatives like Project Maven, which uses machine learning to analyze surveillance data and identify potential targets. Initially a Google partnership, Maven now involves contractors like Palantir and is credited with supporting recent U.S. military operations by enhancing situational awareness for human analysts. The Pentagon's 2026 budget request includes a $13.4 billion standalone allocation for autonomy and AI-driven platforms, signaling a massive increase in investment. This funding prioritizes areas like autonomous sensing, cross-platform data integration, and the application of AI to shipbuilding and maintenance forecasting. Interestingly, the final OpenAI deal reportedly includes the very prohibitions on mass surveillance and autonomous weapons that Anthropic had requested. This has led to speculation that Anthropic's public refusal and the subsequent backlash created the political space for OpenAI to secure a contract with stronger safeguards than might have otherwise been possible.
