OpenAI will require model routers, AI gateways, and similar intermediaries to secure its approval before offering customers direct access to OpenAI models, starting November 20, 2026.
The OpenAI policy announcement draws a line between services that resell or route direct model access and apps that use OpenAI models within their own features. The latter are not covered by this specific approval rule.
Approved partners will also need safeguards aligned with OpenAI’s, processes for investigating suspected violations, and coordination on enforcement. OpenAI is developing SafetyIDs to distinguish end users without learning their identities, but the published text does not impose a blanket SafetyID mandate on every gateway by the deadline.
Gateway operators should review whether their product counts as a model-access intermediary and leave time for approval before the policy takes effect.