Identifying data
ChatGPT bug exposed chat titles and billing data; Italy stepped in
In March 2023 a bug let ChatGPT users see the titles of other people's chats, and let some subscribers' billing details be seen by others. Italy's data protection authority stopped the service; a court later annulled its fine.
On 20 March 2023 a bug caused some ChatGPT users to see the titles of other people's conversations in their own chat history. According to OpenAI, the cause was a bug in an open-source library, redis-py: after certain cancelled requests, the system returned data meant for another user. The company said only the titles were visible, not the content of the conversations.
The bug also affected payment data. During a nine-hour window, some ChatGPT Plus subscribers' names, billing addresses, card types, expiry dates and the last four digits of their cards may have been seen by another user; the full card number was not exposed. According to OpenAI, this may have affected 1.2 per cent of the Plus subscribers active during those nine hours; it notified those affected by email, fixed the bug and added further checks.
At the end of March 2023 Italy's data protection authority, the Garante, citing this breach among other things, imposed an immediate temporary limitation on OpenAI's processing of Italian users' data and opened an inquiry. It also objected that users had not been properly informed, that there was no legal basis for training the models, and that there was no age verification.
At the end of the inquiry, in a decision of November 2024, the authority fined OpenAI €15 million, partly for failing to notify it of the March breach. On 18 March 2026 the Court of Rome annulled that decision. It did not rule on the merits: it held that because OpenAI had meanwhile set up its European establishment in Ireland, the decision fell to the Irish data protection authority.
What can we learn from it?
A provider's bug can expose more than the text of a conversation: even a title says a lot if it contains a client's name or problem. Such bugs are beyond the user's control, but what we type is not. General chatbots work well for professional questions as long as the question identifies no one. That is why Confiella sends only reviewed, pseudonymised text; in copy mode, too, only the confirmed, pseudonymised request goes into your own ChatGPT subscription. When typing by hand, it is enough to leave out or generalise names, identifiers and recognisable details.
Sources
- Garante per la protezione dei dati personali: Artificial intelligence: stop to ChatGPT by the Italian SA. Personal data is collected unlawfully, no age verification system is in place for children
- Help Net Security: A bug revealed ChatGPT users' chat history, personal and billing data
- Il Sole 24 Ore NT+ Diritto: ChatGPT, il Tribunale di Roma annulla la sanzione da 15 milioni del Garante: decide l'autorità capofila
This article is based on the public sources listed. If you find a mistake, write to us: [email protected]