In a significant development for global security, OpenAI has finalized a contract with the Pentagon that includes some of the most stringent ethical safeguards ever placed on military technology. The agreement explicitly forbids the use of OpenAI systems for domestic mass surveillance, the development of fully autonomous lethal weapons, and the use of social credit-style scoring. This deal marks a victory for OpenAI’s “human-in-the-loop” philosophy, ensuring that lethal force remains a human responsibility.
The agreement comes at a time of extreme tension between the tech sector and the federal government. After the Trump administration blacklisted Anthropic for being “uncooperative,” there were fears that the government would seek out a vendor willing to provide “guardrails-off” AI. Instead, OpenAI was able to convince the Department of War that safety and national security are not mutually exclusive, leading to a contract that actually strengthens existing legal protections.
OpenAI CEO Sam Altman noted that the company’s deployment will be restricted to cloud infrastructure, which inherently prevents the software from being used to power autonomous drones or edge-based “killer robots.” Furthermore, OpenAI has requested that the Pentagon offer these same ethical terms to all other AI companies. This move is seen as an attempt by OpenAI to de-escalate the “AI arms race” by establishing a common set of international rules for military engagement.
The technical implementation of the deal involves “safety classifiers” that OpenAI can update in real-time. These tools will automatically detect if a model is being asked to perform a prohibited task, such as tracking American citizens’ private data. By retaining control over its “safety stack,” OpenAI is maintaining a level of oversight that was previously unheard of in major defense contracts.
As OpenAI begins the transition of Pentagon workflows from Anthropic’s systems to its own, the company is also dealing with internal questions from its staff. OpenAI leadership has held a series of memos and meetings to reassure employees that the company is not becoming a “weapons manufacturer.” Rather, OpenAI argues it is providing the necessary tools to ensure that if the military uses AI, it does so in the safest and most responsible way possible.
