OpenAI cuts Mixpanel after third-party breach leaks user names and emails

A security lapse at a data analytics vendor has put a slice of OpenAI’s user base at risk. On December 19, OpenAI updated a prior security disclosure to clarify that a November breach at Mixpanel—its web analytics provider for the API product at platform.openai.com—also affected a limited number of ChatGPT users, not just those on the developer platform. The company is framing this as a clarification, not an escalation, insisting all impacted users were already contacted.

The incident itself unfolded inside Mixpanel’s environment. An attacker gained unauthorized access and exported a dataset containing customer identifiable information. OpenAI was notified on November 25 that the compromised data included user profile details: names, email addresses, coarse location data pulled from browser IPs, operating system and browser info, referring websites, and organization or User IDs. OpenAI is adamant that no chats, API keys, passwords, payment details, or government IDs were exposed. “This was not a breach of OpenAI’s systems,” the company states flatly.

OpenAI’s response was swift and terminal for the vendor relationship. The company removed Mixpanel from its production services and, after its own review, has “terminated its use of Mixpanel” entirely. The move signals zero tolerance, but it’s the post-mortem actions that reveal the real anxiety here. OpenAI says it’s now conducting “additional and expanded security reviews across our vendor ecosystem” and raising the security bar for all partners. That’s the right call, because the breached data—names, emails, and org IDs—is a ready-made kit for precision phishing campaigns. OpenAI’s warning is blunt: treat any unexpected emails with suspicion, verify the sender’s domain, and remember the company will never ask for passwords or API keys over email.

The real sting isn’t the data’s sensitivity, but its weaponizability. An email paired with an accurate name and the knowledge that the recipient uses OpenAI’s API or ChatGPT is a social engineer’s dream. For enterprise admins, the risk compounds if user IDs map to internal org structures. OpenAI is notifying victims directly, but the onus is now on users to stay alert for a wave of highly credible-looking spam that leverages this exact metadata. The breach didn’t crack OpenAI’s vault, but it did hand attackers a key to the front gate of human trust.