OpenAI to require approval for model-access intermediaries
OpenAI is adding a prior-approval requirement for services that give third parties direct access to its models. Applications do not need approval solely because they use the models to power their own features, and the updated Usage Policies take effect November 20, 2026.
OpenAI will require model-access intermediaries to secure its review and approval before giving third parties direct access to OpenAI models. The company announced the change on October 21, and the requirement takes effect November 20, 2026.
The approval gate covers services such as model routers and AI gateways. OpenAI’s updated Usage Policies say the rule applies to a model-access intermediary platform or another service that enables direct third-party access to OpenAI models.
The policy distinguishes those services from ordinary software integrations. An application does not need approval solely because it incorporates OpenAI models to power its own features or functionality. But the published materials do not further define “directly access,” leaving the treatment of services near that boundary unresolved.
OpenAI also said partners providing access to its models must maintain safeguards consistent with its own, investigate potential violations, act when they identify concerns and coordinate enforcement across services. The company said those policies apply to access provided directly by OpenAI, by partners that host its models and by reseller partners. A recent example of the hosting route is AWS offering OpenAI GPT models through Bedrock in Canada. OpenAI described the safeguards as its requirements; the supplied sources do not independently establish that every partner has implemented them.
The intermediary clause appears in the October 21, 2026 revision. OpenAI’s archived October 29, 2025 Usage Policies do not contain the provision. The current policy’s changelog identifies model-access intermediary approval as one of the update’s requirements.
The published materials do not specify how operators should request review, what documentation they must provide, how long approval may take or whether existing intermediaries need new approval or receive a transition period. They also do not provide a technical baseline for safeguards that are consistent with or exceed OpenAI’s own.
More news
OpenAI says research model spent funds from an unauthorized test wallet
FTC investigates OpenAI and Anthropic over AI safety risks
OpenAI engineer charged after hidden recorder found in conference room
