What Happened
On March 20, 2023, users of ChatGPT began reporting that they could see conversation history titles belonging to other users in their sidebar. Investigation revealed that a bug in an open-source library (redis-py) caused requests to be routed to incorrect users under specific conditions. For a subset of affected users, the bug also exposed payment-related information including first and last names, email addresses, payment addresses, and the last four digits of credit card numbers.
The issue affected approximately 1.2% of ChatGPT Plus subscribers who were active during a specific nine-hour window.
Timeline
The bug was introduced during a server-side change on March 20. Users began reporting seeing others’ chat titles within hours. OpenAI took ChatGPT offline to investigate. The service was restored later that day after the bug was patched. On March 24, OpenAI published a detailed post-mortem explaining the root cause and scope of the exposure.
Impact
While the number of affected users was relatively small, the incident was significant as the first major security breach at OpenAI. It exposed the sensitivity of chat history data — users’ conversations with AI often contain highly personal, professional, or confidential information. The idea that other users could see your conversation topics was deeply unsettling.
The incident triggered Italy’s data protection authority (Garante) to temporarily ban ChatGPT in the country, citing privacy concerns. This became the first national ban of a major AI chatbot and prompted other European DPAs to investigate OpenAI’s data handling practices.
Response
OpenAI’s response was relatively transparent. The company published a detailed technical post-mortem identifying the root cause in the redis-py library, explaining exactly what data was exposed and to whom, and outlining the fixes implemented. OpenAI notified affected users and offered to answer questions about the exposure.
The company also implemented additional monitoring and testing to prevent similar issues from recurring.
Lessons Learned
The ChatGPT data breach demonstrated that AI services face the same security challenges as any other web application, with the added sensitivity that AI conversation data often contains more personal and confidential information than typical web service data. Users share things with AI chatbots that they might not share with other services.
The incident also showed how a relatively minor technical bug (a race condition in a caching library) could have outsized consequences when it affects a service processing highly sensitive conversational data at massive scale.