Starting Nov. 20, operators of intermediary services that let other users access OpenAI models directly will need the company’s review and approval.
The company published the change to its Usage Policies Oct. 21. The approval requirement is triggered by providing direct access to others; an app that integrates a model into its own functionality does not need approval on that basis alone. That exception is limited to the approval rule and does not remove an app's obligations under the rest of the Usage Policies.
OpenAI names model routers and AI gateways as examples of services covered by the review process. Providers authorized to offer access must apply protections that meet OpenAI's standards, examine suspected misuse, address concerns and align enforcement across their services.
The company says the revision is intended to keep safeguards consistent among services through which people reach its models. Other parts of the update address using OpenAI models or their outputs to train alternative models, protections for children and teens, human review in high-stakes settings, and government work involving national security, law enforcement and intelligence.
The new review and approval requirement takes effect Nov. 20.
