Pentagon Approves OpenAI AI Safety Framework After Anthropic Split

The Pentagon has formally adopted OpenAI’s safety guidelines for artificial intelligence, marking a significant shift in its approach to AI oversight after ending its partnership with Anthropic. The move, first reported by Axios, positions OpenAI as a primary partner in the Department of Defense’s ongoing efforts to set boundaries for responsible and safe AI deployment.

Pentagon Shifts AI Vendors Amid Safety Concerns

The approval of OpenAI’s safety red lines comes just hours after the Trump administration enacted a ban on Anthropic, a prominent AI vendor previously engaged with the Pentagon. According to Axios, this decision follows internal concerns about Anthropic’s adherence to evolving government safety standards. The Pentagon’s selection of OpenAI reflects its ongoing drive to ensure that AI systems used in defense applications align with official responsible AI initiatives and safety protocols.

• The Pentagon’s Responsible AI Strategy focuses on transparency, accountability, and human oversight in military AI systems.
• OpenAI’s safety red lines set explicit limits on AI system autonomy, decision-making, and use in lethal applications.
• The shift away from Anthropic was catalyzed by new compliance requirements under federal law.

What the OpenAI Guidelines Include

OpenAI’s safety framework, now formally adopted by the Department of Defense, outlines restrictions on the use of advanced AI in scenarios that could pose risks to human life or violate international norms. These guidelines dovetail with the Pentagon’s published Responsible AI Strategy and Implementation Pathway, which includes:

• Mandatory human review of AI decisions in critical missions
• Prohibitions on AI systems making lethal decisions independently
• Strict auditing, testing, and red-teaming of AI models prior to deployment
• Transparency requirements for vendors regarding system capabilities and limitations

This alignment with the NIST AI Risk Management Framework—frequently referenced in federal AI procurement—demonstrates the Pentagon’s intent to bring civilian AI safety standards into the defense sector.

Regulatory and Legislative Context

The decision to replace Anthropic with OpenAI comes in the wake of updated requirements under the National Defense Authorization Act for Fiscal Year 2022, which mandates stricter controls on AI vendor contracts and more robust safety guarantees. These requirements seek to operationalize principles from the Blueprint for an AI Bill of Rights, ensuring government-funded AI systems are transparent, fair, and accountable.

Implications for the Defense Industry

The Pentagon’s pivot to OpenAI as a preferred partner for AI safety sets a precedent for future government contracts. Federal contract data on Sam.gov shows a growing emphasis on explicit safety and compliance deliverables in AI-related awards. This trend is likely to shape how other vendors design and propose their AI solutions for defense use.

Looking Forward

With OpenAI’s guidelines now in place, the Pentagon is expected to increase oversight of AI deployments, leveraging lessons from the Anthropic split. As adoption of AI in military contexts accelerates, continued scrutiny and alignment with civilian safety standards will remain central to future government partnerships. The shift highlights the importance of clear, enforceable red lines for AI use—a principle that other agencies and international partners are likely to follow as the technology evolves.