Hundreds of private conversations between users and the AI chatbot Claude were found to be publicly accessible through Google searches, according to reports. The discovery came after Reddit users noticed that using a specific site-search term on search engines revealed dozens of chat logs.
How the Chats Were Discovered
Reddit users unearthed the conversations by employing a site-specific operator in search queries, which limited results to particular domains. A repository on GitHub indicated that some 755 messages were searchable across multiple search results pages. The chats Metro reviewed included a range of topics, from the absurd to the personal.
Examples of the Conversations
One user asked Claude whether birds are “secretly professional rappers who co-wrote most of Jay Z’s albums,” adding that they could try to find the bird and interview it but thought it might be hard because “it’s clearly a celebrity.” Another sought advice on how to transform into a nine-tailed fox, a spirit from Japanese mythology. A third requested to be called “crude” terms that could not be published. Other chats included Claude fact-checking a user who denied human-caused climate change, helping someone invent the word “transtill,” and playing cards (which Claude won). In a nearly 20,000-word conversation, a user asked Claude if it is “situationally aware” and later told the algorithm they “feel bad” for it.
How the Chats Ended Up on Google
Claude, developed by Anthropic, allows users to share conversations via unique URLs. Once shared, anyone with the link can view the chat snapshot. Anthropic stated that these shareable links are “not guessable or discoverable” and that the company does not share chat directories with search engines. However, if users post these links on forums or social media, Google’s web crawlers may index them. Inside each chat log’s code, a robots.txt file instructs search engines to avoid indexing, but this is not always effective. Metro found that some of the previously Google-able logs lacked a “noindex” tag, which is recommended by Google for preventing indexing.
Company and Search Engine Responses
Anthropic told Metro that they give users control over sharing their conversations and adhere to privacy principles. Google responded by noting that it does not control what pages are made public on the web and that these pages were indexed across many search engines. They emphasized that site owners have clear controls to decide whether pages can be crawled or indexed. By Tuesday evening, a test search by Metro on Google, Yahoo, Bing, DuckDuckGo, and Brave returned no results using the same trick.
Similar Incidents with Other AI Chatbots
This issue echoes previous events with other chatbots. Last year, 4,500 “shared” ChatGPT chat logs were found to be accessible via Google, prompting OpenAI to tighten data controls. More recently, around 300,000 chat logs from Elon Musk’s Grok AI on platform X were also viewable through a quick Google search.
