OpenAI requires review and approval for gateways and routers offering its models
For the head of third-party risk at a regional bank
OpenAI updated its Usage Policies effective Nov. 20 to require model routers and AI gateways that provide direct access to its models to get OpenAI's review and approval first.2
OpenAI said in an Oct. 21 post that it worked with customers and partners over the past few months to align safety policies across services that provide its models, and wants safeguards applied consistently whether people use its models directly, through hosting partners or through resellers.2 Semafor reported that US security agencies and companies, including Anthropic, have accused Chinese AI labs of distilling, or copying, American frontier models by routing requests through resellers that obscure the end user's identity; OpenAI's post does not mention distillation or China.12
According to Semafor, those cases largely involved gray-market resellers known in China as "transfer stations," but the new rule extends OpenAI's oversight to any service between it and the people using its models.1
OpenAI's policies also require partners that provide access to its models to maintain safeguards consistent with its own, investigate potential violations, and take appropriate action when concerns are identified.2
OpenAI said it is working with customers on SafetyIDs, which let it tell individual end users apart in an investigation without learning their identities.2
OpenAI said that when it restricts access, it revisits the restriction if meaningful changes in conduct, commitments or safeguards address the concerns behind its decision; it did not publish approval criteria or timelines in the post.2
Banks and fintechs that reach OpenAI models through a router or gateway now depend on that intermediary holding OpenAI's approval, which is a new vendor-risk and continuity question to put in contracts and due diligence. Teams should ask router vendors about their approval status and line up a fallback path.
Our take. Your vendor inventory probably lists the AI tool, not the router behind it. OpenAI now decides whether that router keeps access, so the contract that matters most may be one your bank never signed.
Action. Third-party risk: List every vendor and internal tool reaching OpenAI models through a router, and confirm that router has OpenAI approval.
What our records say
ChatGPT Enterprise is OpenAI's workplace version of ChatGPT, launched in 2023, and it sells banks ChatGPT seats for employees and API access to its models.1 ChatGPT Enterprise is OpenAI's workplace version of ChatGPT, launched in 2023, and it sells banks ChatGPT seats for employees and API access to its models; in 2026 it added on-site engineering deals, a financial-services edition built with investment-bank design partners, and a separately funded deployment company.1
From our records
- ChatGPT Enterprise (OpenAI)vendor · banks using it (17) Morgan Stanley, BNY, Customers Bank · alternatives Microsoft Copilot, Claude, Gemini Enterprise
- Claude (Anthropic)vendor · banks using it (24) Goldman, JPMorgan, BankUnited · alternatives Microsoft Copilot, ChatGPT Enterprise, Gemini Enterprise
Sources
- ReportOpenAI tightens grip on model resellers | Semafor Semafor
- Primary sourceBroad access through consistent standards | OpenAI OpenAI