Watchtower

OpenAI updated its Usage policies on Oct. 21, with the changes going into effect Nov. 20. The largest changes were to its policies on third parties using OpenAI models and their outputs to train other AI models. 

Broadly, the changes were to prohibit it. Users may not use OpenAI models or their outputs to generate or select training material for models intended as alternatives to OpenAI models, including generating training data, providing feedback, or validating data. Attempting to extract protected reasoning “through unauthorized access, reverse engineering, or circumvention of safeguards” is prohibited, and restrictions apply when outputs from normal use are later used for training purposes.

Other changes include a new section describing “OpenAI-approved use,” saying that OpenAI may grant specific partners exceptions to certain restrictions in its Usage policies. It presents “national security, intelligence, and law enforcement” as an example, although it says all customers must comply with requirements on preserving human judgment and accountability over high stakes decisions.

The definition of prohibited high stakes actions in sensitive areas without human review was clarified, including “actions, decisions, transactions, submissions, and other steps that materially affect a person’s rights, safety, access to essential services, or other consequential interests” in relevant domains.

There were multiple changes to the policy; this is not an exhaustive list.

A diff of the changes can be found below: