The Pulse
OpenAI Requires Approval for Model Routers and AI Gateways
OpenAI says model routers, AI gateways and other intermediaries must receive prior review and approval before giving users direct access to its models. The updated Usage Policies take effect November 20, 2026, while partners must maintain s
AI.info Team · Written with AI · fact-checked ·
OpenAI says developers and businesses should be free to choose how they access its models. But starting November 20, services that route or gateway users directly to those models will need the company’s review and approval first—a new condition on the very choice OpenAI says it wants to preserve.
The company announced the policy changes on October 21, alongside new expectations for partners that host or resell its models. OpenAI’s stated aim is to keep safeguards consistent wherever customers get access. The announcement does not name affected providers or explain how the approval process will work.
Approval applies to direct-access intermediaries
OpenAI’s updated Usage Policies require any “model-access intermediary platform or other service” that enables third parties to access its models directly to obtain prior review and approval. The company names model routers and AI gateways as examples. The policies make a distinction for applications that integrate OpenAI models to power their own features: integration alone does not trigger the approval requirement.
That line puts the focus on services acting as a route to the model, rather than every product built with one. OpenAI does not spell out how it will classify borderline services, what information applicants must provide, or how long review will take. The policy page says the changes take effect November 20, 2026.
Partners must investigate and coordinate
The new approval rule sits alongside broader requirements for partners that host or resell OpenAI models. They must maintain safeguards consistent with OpenAI’s own, investigate potential violations and take appropriate action when concerns arise. OpenAI also says partners need to coordinate enforcement so restrictions remain effective across services.
That requirement addresses a practical complication of distribution: a restriction on one service may not be enough if a user can reach the same model through another. OpenAI says it is developing the standards with partners and will account for differences in how services operate and how customers use them. The company does not publish specific technical controls or say how it will audit partner compliance.
OpenAI frames oversight as a condition of choice
OpenAI presents the rules as a way to preserve access through different providers while setting a shared safety baseline. “Developers and businesses should be able to choose where they access our models, with consistent safeguards across those services,” the company writes. Its argument is that distribution can remain broad if providers enforce comparable protections.
The approval requirement nonetheless gives OpenAI a gatekeeping role over third-party services that offer direct access. The company says its enforcement decisions will rely on documented conduct and risk, and that it will revisit restrictions when changes in conduct, commitments or safeguards address the concerns behind them. It also says it is working with customers on SafetyIDs, identifiers intended to help distinguish end users during misuse investigations without revealing their identities to OpenAI.
November 20 is the policy deadline
The new Usage Policies take effect on November 20, giving intermediaries a defined date to meet the review requirement. OpenAI says customers and partners will be able to ask questions or raise compliance concerns, but it has not described a review timetable or appeal process specific to model-access providers. For now, the policy sets the obligation and the effective date; the announcement leaves the mechanics of approval open.