OpenAI updates usage policy, effective Nov 20; model routers and AI gateways need approval
In short
OpenAI is updating its usage policy to clarify rules on safe model development and sensitive AI uses, effective 20 November 2026. According to a statement cited by Chinese outlets, model routers and AI gateways must be reviewed and approved by OpenAI before offering direct model access.
Read the full story 1 min read
OpenAI is updating its usage policy, with the new version taking effect on 20 November 2026, CLS reported on 22 October. The update is meant to clarify rules on safe model development and on sensitive uses of artificial intelligence. [ 1 , 2 , 3 ]
According to a statement cited by 36Kr and Jiemian, the policy requires model routers and AI gateways to be reviewed and approved by OpenAI before they provide direct model access. Partners must adopt safeguards consistent with OpenAI's standards, investigate violations and coordinate enforcement. The 36Kr text credits Jiemian, and Jiemian dates its report 21 October 2026. [ 2 , 3 ]
Why it matters
The documents describe a rule change for partners that offer access to OpenAI models, including model routers and AI gateways. They give no further detail on how many companies are affected or how the review will work.
Key facts
- OpenAI is updating its usage policy, effective 20 November 2026. [ 1 , 2 , 3 ]
- The update is meant to clarify rules on safe model development and sensitive uses of AI. [ 1 , 2 , 3 ]
- Model routers and AI gateways must be reviewed and approved by OpenAI before providing direct model access. [ 2 , 3 ]
- Partners must adopt safeguards consistent with OpenAI's standards, investigate violations and coordinate enforcement. [ 2 , 3 ]
Confirmed by several sources
Still unclear
- How the review and approval process for model routers and AI gateways will work. The documents are short summaries and do not describe the process or criteria.
- What the policy defines as sensitive uses of AI. CLS mentions sensitive uses only in general terms and gives no specifics.
- Under the policy, model routers and AI gateways need OpenAI review and approval before providing direct model access. Reported by a single source so far