OpenAI now requires approval to run an AI gateway or model router
Updated usage policies effective today require OpenAI sign-off for any service that gives third parties direct access to its models, though apps that just call the API internally are unaffected.
OpenAI’s updated Usage Policies take effect today, 20 November, having been announced on 21 October. The headline change: anyone operating a model-access intermediary, a service whose product is giving third parties direct access to OpenAI’s models, now needs OpenAI’s prior review and approval to launch or keep running it.
The requirement is aimed at model routers and AI gateways, services that exist to resell or forward raw access to OpenAI’s models, and explicitly excludes apps that simply integrate a model into their own product’s features. The same update bars using OpenAI’s outputs to build training data, reinforcement-learning environments or scored feedback for a competing model, and expands safeguards for users under 18.
Why it matters: if you run, or are building, a multi-model gateway, a bring-your-own-key proxy, or anything that routes end users’ requests straight through to OpenAI’s API under your own service, you now need OpenAI’s sign-off, whether the service is new or already live. Products that just call the API inside their own app are not affected.
The caveat: OpenAI’s policy states the approval requirement without publishing a process, timeline or contact point for operators of intermediaries already running.