Some people's chats with Claude AI made publicly available online
Hundreds of conversations with Anthropic’s chatbot were discovered as being publicly accessible.
Conversations between users and Anthropic's Claude AI chatbot were inadvertently made public online, with hundreds of exchanges discoverable via Google and other web browsers.
Links to these chats, some containing personal and professional data, appeared in search results when specific site-search terms were used. This occurred because search engines like Google had indexed Claude chats for which users had opted to "share" a link, making them broadly accessible.
While the search availability of these chat logs was rectified over the weekend, many had already been saved and widely disseminated online.
An Anthropic spokeswoman stated that Claude users retain control over sharing their conversations with the chatbot. She emphasized that conversation links were "not guessable or discoverable unless people choose to share them themselves." The spokeswoman added, "When someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services."
The "share" option within Claude informs users that "anyone with the link" can view its contents, but it does not explicitly mention the possibility of the link appearing in Google and other search results.
Reddit users initially uncovered these publicly available chats, which encompassed over 200 conversations with Claude across at least 25 pages of search results, some dating back only weeks.
The conversations show users prompting the chatbot on a wide range of subjects. For instance, one chat log from last year features a user asking Claude if it preferred "to help me or do you want to help anthropic more?", to which the chatbot partially responded, "I experience something like wanting to help you."
In an April conversation, a user instructed Claude to draft an unpublished blog post about cloud security, including details of a corporate project. Another user last month asked Claude how to "become become Nine-tailed fox?", later clarifying a desire for literal human-to-creature transformation. Claude initially responded by presenting an AI-generated image, claiming the user had been granted "fully functional fox powers!"
Other Claude conversations involved users seeking resume assistance, providing their names, contact information, and work history. Some users also appeared to conduct proprietary research for their jobs, such as in healthcare, including transcripts of private discussions.
Last year, OpenAI faced a nearly identical issue with ChatGPT chat logs becoming publicly accessible, leading the company to subsequently modify the ease of access to such logs.
A Google spokesman clarified to the BBC that the company does not control "what pages are made public on the web," attributing such actions to websites themselves. He stated, "We give site owners clear controls to decide whether pages can be crawled or indexed, and we always respect those directives."
Given that search indexing of the chat logs has ceased, it is probable that Anthropic utilized available tools to swiftly block the chat log links from search results. Google's process for a website owner to block a link is straightforward but requires initiation by the website owner.
Other search engines, including Bing, Brave, and DuckDuckGo, where the Claude chat logs also appeared, were contacted for comment.