What happened
OpenAI published “Broad access through consistent standards” on 21 October 2026. It says that over the past few months it aligned safety policies across the services that provide its models, and that it is updating its Usage Policies accordingly.
Key details
All details are from OpenAI’s post.
- New requirement: the Usage Policies update, “effective November 20, 2026”, requires services that provide direct access to OpenAI’s models, “such as model routers or AI gateways”, to get review and approval first.
- Who is covered: OpenAI says the policies apply to direct access to its models, to partners that host them and to reseller partners.
- SafetyIDs: customers can supply SafetyIDs to distinguish their end users in safety investigations without OpenAI learning those users’ identities.
- Private Safety Processing: the post cites it as an example of customers keeping control over sensitive data.
- Not stated: the post does not describe the review process, timelines for approval or which existing services are affected, and it gives no pricing or model specifications.
Why it matters
Developers who reach OpenAI models through a router, gateway or reseller, and the companies running those services, should check how the new approval step applies to them before 20 November. The detail that decides the impact, how review works, is not in OpenAI’s post and is likely to be in the updated Usage Policies, which we did not compare line by line.
Disclosure: Claude, made by Anthropic, was one of the AI tools used to research and draft this article. Details come from OpenAI’s post of 21 October 2026, read on 22 October 2026. OpenAI competes with Anthropic.
Sources
- Broad access through consistent standards Primary source , OpenAI