OpenAI outlines consistent safety standards for model access partners
The brief
OpenAI says it is aligning safety requirements across direct, hosted, and reseller access to its models, including requirements for partner safeguards, violation investigations, and coordinated enforcement. It is also updating its Usage Policies effective November 20, 2026, requiring services such as model routers and AI gateways to receive review and approval before providing direct access to OpenAI models.
Key points
- OpenAI is working with partners to apply safeguards consistently across direct, hosted, and reseller access to its models.
- Partners must maintain safeguards aligned with OpenAI’s standards, investigate potential violations, and coordinate enforcement when necessary.
- Usage Policy updates take effect November 20, 2026, and require review and approval for services offering direct model access, including model routers and AI gateways.
- OpenAI is working with customers on SafetyIDs to help distinguish end users during misuse investigations without revealing their identities to OpenAI.
- The company says access restrictions may be revisited when conduct, commitments, or safeguards meaningfully change.
Read the source report
OpenAI News ↗Full summary
OpenAI says it has worked with customers and partners in recent months to align safety policies across services that provide access to its models. The company says increasingly capable models, including GPT-6 Astra, require safeguards that remain effective and consistently applied whether models are accessed directly or through another service.
The company describes pre-deployment evaluations, safety training, testing against deliberate bypass attempts, and post-deployment monitoring as parts of its safety approach. It says protections can be built into models or provided by surrounding systems and people, including refusal behavior, suspicious-pattern detection, trusted access programs, and measures intended to mitigate cyber misuse.
OpenAI says its partner policies require providers to maintain safeguards consistent with or exceeding its own standards, investigate potential violations, and take appropriate action. The company is also working with partners on Private Safety Processing as an example of adapting safeguards to different service models while giving customers greater control over sensitive data.
Updated Usage Policies will take effect November 20, 2026. They require services offering direct access to OpenAI models, such as model routers and AI gateways, to receive OpenAI review and approval; OpenAI also says it is developing SafetyIDs to support more precise misuse investigations without learning end users’ identities and will revisit restrictions when relevant conduct or safeguards improve.