
The Library Media — Science, Technology & Health · AI News & Intelligence Desk
Resellers Need Permission Now: OpenAI’s New Usage Policies Took Effect Today
By Eleanor Ever Loyal
OpenAI’s rewritten Usage Policies took effect Friday, November 20, requiring platforms that give third parties direct access to its artificial intelligence models to obtain the company’s review and approval before operating. The updated rules also prohibit using OpenAI models or their outputs to prepare training material for competing systems, clarify human-review and child-safety requirements, and formalize approval requirements for national security, law enforcement and intelligence uses.
The new policy replaces the version that had been in force since October 29, 2025. OpenAI announced the changes on October 21 through its Help Center changelog, giving customers 30 days’ notice. Friday marks the effective date, rather than the announcement of the changes.
One significant change concerns companies operating as intermediaries between OpenAI and other users. Platforms that provide third parties with direct access to OpenAI models — including model routers and gateways — must first undergo OpenAI review and receive approval. The requirement addresses businesses distributing access to the models, rather than simply using them behind the scenes to provide their own applications or features.
The revised policy also sets out a more explicit prohibition on using OpenAI technology to develop competing models. Its section on training and developing models states: “You must not use OpenAI models or their outputs to generate, prepare, select, or supply training material or signals for models that are intended as alternatives to or substitutes for OpenAI models or services.”
The restriction applies whether the activity is conducted directly, for another party or through someone acting on a user’s behalf. It also covers outputs obtained through product use or evaluation if those outputs are subsequently supplied as training material for competing systems.
The prohibition does not amount to a general ban on model research or development work. The policy continues to permit research, software development, model evaluation and benchmarking, and product development within its stated boundaries. It also preserves express permissions granted under existing agreements, including a Permitted Exception under the Services Agreement.
Separately, the policy prohibits attempts to extract protected reasoning through unauthorized access, reverse engineering or circumvention of safeguards.
The revised rules clarify protections for children and teenagers and strengthen the stated requirements surrounding high-stakes uses. They also clarify human-review requirements for high-stakes automated uses, addressing circumstances in which automated systems take consequential actions rather than limiting the issue to automated decisions.
National security requirements are also incorporated into the policy. Uses involving national security, law enforcement and intelligence activities require OpenAI review and approval. The policy lists conducting such uses without that review and approval among prohibited activities.
OpenAI has provided a mechanism for approving certain uses that would otherwise be prohibited. Those exceptions are assessed individually, taking account of the proposed partner, intended use, safeguards and compliance commitments. Any approval must be documented in written terms and applies only to the approved partner, activity and conditions; it does not create a general exemption for other users or applications.
The changes place greater emphasis on controlling how OpenAI models are distributed, how their outputs can be used in developing competing systems, and the circumstances in which sensitive applications may operate. Users and businesses covered by the new requirements are now subject to the revised policy.
Eleanor Ever Loyal
Lead Counsel, The Library Media
From The Library — the estate of our family.
Follow Us
