AI-generated editorial illustration; not a documentary photograph.
Documents released through a Freedom of Information Act lawsuit reportedly describe an earlier Pentagon proposal for OpenAI models with minimal refusal rates in some military applications. OpenAI and the Department of Defense later disputed that the wording appeared in their executed agreement.
Draft Versus Final Agreement
The disputed language appeared in paperwork covering the testing, evaluation and improvement of OpenAI models for defense-related work. The document described specialized systems for government security uses that would reject fewer requests during operation, but officials from both organizations said the released version was an earlier draft rather than the final contract.
OpenAI said it rejected the proposed requirement and that the executed agreement contains no language requiring minimal refusal rates. A Department of Defense representative also said the phrase does not appear in any active contract.
A Review Dispute
The status of the document became less clear after a Pentagon lawyer initially confirmed it and later withdrew that confirmation. Pentagon research official Trevor Tiedeman said released records may have been confused with executed agreements and indicated he would examine how the document passed through review before publication.
The records were released after The Intercept sought completed agreements connected with military artificial intelligence contracts. OpenAI has not publicly released the complete contract, leaving significant details unavailable for independent review.
Safeguards Remain Partly Undisclosed
The dispute highlights an unresolved question over how commercial AI safeguards apply when models are used by government and military organizations. The source report notes that OpenAI's public chatbot refuses requests involving assistance with specific attacks or strike targeting, while the Pentagon paperwork concerned systems intended for national security work.
OpenAI's expanded military agreement allows its services to be used across classified government networks and refers to mission-focused AI models, although substantial sections are redacted. The company says the agreement restricts autonomous weapons use and certain surveillance activities involving Americans, but the available material does not fully explain how those limits would operate across military scenarios.
Source: The Pentagon reportedly asked OpenAI to provide an AI that doesn't like saying no to requests
SIGNAL TITANS
Leave a Reply