OpenAI to require approval for model routers from Nov 20

What changed?

On Wednesday OpenAI said (OpenAI) it is aligning safety policies across every service that offers its models. The rules apply whether people use OpenAI directly, go through partners that host the models, or buy through reseller partners.

The most concrete change: updated Usage Policies take effect on November 20, 2026. From then on, services that provide direct access to OpenAI models, such as model routers or AI gateways, need OpenAI's review and approval before offering that access.

What partners must do

• Maintain safeguards consistent with OpenAI's own.
• Investigate potential violations and take appropriate action when concerns come up.
• Coordinate on enforcement, so a restriction cannot simply be bypassed through another provider.

OpenAI says it has spent the past few months on this with customers and partners, checking that each implementation meets or exceeds its own safety standards. It cites its work on Private Safety Processing as one example, which gives customers more choice and control over sensitive data.

SafetyIDs for misuse cases

A new piece is SafetyIDs. Customers can provide them so OpenAI can tell individual end users apart during an investigation without learning who they are. The aim is to respond to misuse more precisely and with less disruption to others. OpenAI names compromised accounts and leaked API keys as typical sources of misuse.

On enforcement, OpenAI says decisions rest on documented conduct and risk and are proportionate. Restrictions are revisited when meaningful changes in conduct, commitments or safeguards address the original concern.

What it means for developers

If you call OpenAI models through a router or gateway rather than directly, check before November 20 whether your provider is handling the approval. If your app serves many end users on one API key, SafetyIDs are worth a look too, since they make it less likely that one bad user takes the whole account down with them.