OpenAI Sets Common Safety Rules for Model Access Providers

OpenAI is updating its Usage Policies to clarify safety requirements across services that provide its models, as detailed on the company website. The rules take effect on November 20, 2026, and require model routers, AI gateways, and similar services to receive OpenAI approval before offering direct model access.

Partners that host or resell OpenAI models must maintain safeguards consistent with the company's standards. They must also investigate potential violations, take action when concerns arise, and coordinate enforcement so restrictions remain effective across services.

OpenAI is also developing SafetyIDs, which customers can provide during misuse investigations. The identifiers are designed to distinguish individual end users without disclosing their identities to OpenAI, allowing restrictions to target specific misuse while limiting disruption to other users.