The Pentagon sought AI chatbots from OpenAI with minimal refusal rates, as revealed by contracts showing military ties with AI labs.

What it means

The Intercept reports that the Pentagon sought to acquire AI chatbots from OpenAI and other labs with 'minimal refusal rates'—meaning the models would be less likely to decline to answer queries that violate typical AI safety rules, such as requests for harmful or military-applicable information. The article draws on contracts that reveal deepening ties between AI companies and the U.S. military, raising concerns about the weaponization of AI.

Accuracy & sourcesGrounded

The claim is based on contractual documents obtained by The Intercept, a reputable investigative news outlet. As a secondary report citing primary sources, the accuracy depends on the veracity of those documents, which have not been independently verified by other outlets in this summary. The reporting appears to be well-sourced and specific.

How new

Genuinely new: this investigation was published on September 8, 2026, and reveals previously undisclosed contract details.

Go deeper

• The Intercept (main)
• The Intercept (Pentagon contract)