OpenAI halts ChatGPT chat indexing by Google, privacy restored
OpenAI has addressed concerns regarding the privacy of ChatGPT conversations after it was discovered that publicly shared chats were being indexed by search engines like Google, Bing, and DuckDuckGo. The company has now removed the feature responsible for this discoverability.
The issue stemmed from ChatGPT’s “Share” feature, which allows users to generate a unique link to their conversations with the AI chatbot, enabling them to share these interactions with others. While conversations are not public by default, those shared using this feature were inadvertently being indexed by search engines. This meant that a simple search query, such as “site:chatgpt.com/share,” could reveal these exchanges.
Among the indexed conversations, many were benign. However, some contained highly sensitive and personal information, including discussions about mental health, sex life, career advice, addiction, and physical abuse. Fast Company, the publication that initially reported on the issue, identified approximately 4,500 conversations visible on Google’s site search. They noted that the actual number of indexed chats was likely higher and that some could contain personally identifiable information if users had included it in their prompts. A Google spokesperson confirmed that OpenAI was responsible for the visibility of these chats to search engines.
An OpenAI employee announced on X (formerly Twitter) that the company had removed the feature, describing it as a “short-lived experiment.” The employee stated that it “introduced too many opportunities for folks to accidentally share things they didn’t intend to.”
OpenAI’s Shared Links FAQ previously indicated that chats would only be made discoverable if users explicitly ticked a “Make this chat discoverable” option. However, it appears many users may have inadvertently selected this option, potentially misunderstanding its implications for public visibility.
Users retain the ability to manage and delete previously shared links through their ChatGPT settings under the ‘Data controls’ section. It is important to note, however, that even if a link is deleted, it might temporarily persist on cached pages of search engines.
This incident also highlights previous statements by OpenAI CEO Sam Altman, who has cautioned that users should not expect legal confidentiality for their ChatGPT conversations. Altman cited the current lack of a comprehensive legal or policy framework for AI, suggesting that OpenAI might be compelled to disclose user chats in a court of law if legally required.