OpenAI to require approval for model routers and AI gateways under updated Usage Policies
OpenAI announced on October 21, 2026 that it is updating its Usage Policies to require services that provide direct access to its models - specifically naming model routers and AI gateways - to obtain OpenAI's review and approval before offering that access. The updated policies take effect on November 20, 2026.
The change is part of a broader push, outlined in a post titled 'Broad access through consistent standards,' to keep safety safeguards consistent no matter where customers access OpenAI models. As models such as GPT-6 Astra expand what developers can build, OpenAI says protections must hold whether access comes directly or through a partner.
Model routers and AI gateways sit between applications and model providers, directing traffic across models and vendors. Under the new policy, any such service offering direct access to OpenAI models needs explicit approval, giving OpenAI a compliance checkpoint over intermediaries that previously operated without one.
For enterprises, the near-term action is inventory: if your stack routes OpenAI traffic through a third-party gateway or router, confirm with that vendor whether it has or is seeking OpenAI's approval before the November 20 effective date. Unapproved intermediaries could become a policy violation risk even if your own usage is compliant.
The policy also clarifies enforcement expectations. OpenAI says it will keep explaining what is allowed and restricted, and describes its misuse response as collaborative - working with customers on issues like compromised accounts or leaked API keys rather than immediate cutoff.
The bigger picture: model access is professionalizing into a permitted-channel model similar to other platform ecosystems. Buyers should expect approval status to become a standard procurement question for AI infrastructure vendors, and should watch for how OpenAI applies the same standards to its largest cloud partners.