API users urged to stay vigilant after third-party data breach

OpenAI has begun warning its API customers after confirming that a security incident at analytics provider Mixpanel exposed limited account information linked to its developer platform. The company says the breach occurred entirely within Mixpanel’s systems, but it has still moved to notify affected organisations, admins and users directly and to cut ties with the vendor in production services.

According to OpenAI’s incident notice, Mixpanel detected unauthorised access to part of its environment on 9 November 2025, when an attacker exported a dataset containing customer-identifiable information and analytics data. Mixpanel later shared the subset relating to OpenAI with the company on 25 November, enabling OpenAI to review what was exposed and begin a wider security response across its vendor ecosystem.

The warning emails stress that there was no breach of OpenAI’s own infrastructure. Core systems, chat content, API prompts and responses, usage data, passwords, API keys, payment details and government IDs all remain unaffected. The incident was limited to analytics data collected from the frontend of platform.openai.com, the interface used by developers and organisations that access OpenAI’s models via API.

Information potentially exposed includes names provided on API accounts, email addresses, approximate location data based on browser metadata, device details such as operating system and browser type, referring websites and organisation or user IDs associated with those accounts. While this data is less sensitive than credentials or financial information, it still offers a useful starting point for phishing and social-engineering attempts targeting technical teams.

In its communication, OpenAI is urging API customers to be on guard for suspicious emails or messages that reference their use of the platform. The company reiterates that it will never ask for passwords, API keys, one-time passcodes or other sensitive details over email, text or chat, and that users should carefully verify sender domains before responding to any unexpected security alerts or account notices.

As part of its response, OpenAI has removed Mixpanel from all production services and is conducting additional security reviews across its third-party vendors. The firm says it plans to raise security requirements for external partners, tightening how user information can be collected and stored, and reviewing whether similar analytics datasets are necessary at all. The move reflects growing scrutiny of how AI providers manage data flows to supporting tools.

For the vast majority of ChatGPT users, OpenAI has confirmed there is no impact from this incident. The affected dataset relates specifically to the API product, rather than consumer-facing chat services. However, the episode has reignited broader questions about how much identifiable information large AI companies share with outside analytics platforms and what safeguards they apply around that practice.

Security experts note that even when passwords and secrets are not exposed, clean lists of names, email addresses and organisation identifiers can be highly valuable to attackers. Such data can help craft convincing spear-phishing campaigns aimed at engineers or administrators who manage API keys and access controls, turning an indirect breach into a more serious compromise if vigilance slips.

In the short term, OpenAI is advising API customers to review their account security hygiene. Recommended steps include enabling multi-factor authentication wherever possible, confirming that administrative email addresses are up to date and controlled, and reminding staff not to reuse passwords across services. While the company says password resets or API key rotations are not required, some organisations may choose to take extra precautions as a matter of policy.

Longer term, the Mixpanel incident is likely to fuel calls for tighter rules around data minimisation and third-party risk in the AI sector. As more businesses rely on OpenAI’s models for customer support, coding tools and internal workflows, the expectation will be that any non-essential sharing of identifiable information is curtailed. For now, OpenAI’s warning to API customers underscores that even limited analytics leaks can have real-world consequences if they are not handled quickly and transparently.