Hundreds of chats with Anthropic's Claude chatbot were left publicly accessible on search engines like Google

Hundreds of conversations that various users had with Claude, Anthropic's artificial intelligence chatbot, turned out to be accessible to anyone on the internet through search engines such as Google. The problem arose because Claude lets users generate a link to "share" a conversation, and some of those links ended up being crawled and indexed by search engines, so that they appeared when using search terms specific to that site. According to the BBC, more than 200 different conversations were found across at least 25 pages of search results, some of them from just a few weeks ago.

The availability of these chats in search results was removed over the weekend, but by then many conversations had already been saved and widely shared across the internet. The discovery was initially made by Reddit users, who noticed that links shared from Claude could be found through searches targeting that specific domain.

An Anthropic spokeswoman explained that Claude users retain control over whether or not they share their conversations, and that links to those conversations "are not guessable or discoverable unless people choose to share them themselves." She added that "when someone shares a conversation, they are making that content publicly accessible and, like any other public content on the web, it can be archived by third-party services." However, the sharing option within Claude only warns the user that "anyone with the link" will be able to see the content, without explicitly stating that the link could end up appearing on Google or other search engines.

The exposed conversations included both personal and work-related information. One user, a year ago, asked Claude whether it wanted to "help me or help Anthropic more," to which the chatbot replied, in part, that "I experience something like wanting to help you." In another conversation from April, a user asked Claude to draft an unpublished blog post about cloud security that included details of a corporate project. In another, from last month, a user asked Claude how to "become a nine-tailed fox," clarifying that they wanted to literally transform from a human into that creature; Claude responded by trying to show them an AI-generated image in which it claimed they had been granted "fully functional fox powers."

Other leaked conversations included users asking for help writing their résumés, with their names, contact details and employment history visible. Some users had even used the chatbot for what appeared to be confidential work-related research, for example in the healthcare sector, including transcripts of private conversations.

This is not an isolated incident in the industry. Last year, OpenAI suffered a virtually identical problem with ChatGPT conversation logs that became publicly accessible, after which the company ended up changing how easily those logs could be found. Similarly, Grok, the AI chatbot built into the social network X, owned by Elon Musk, also saw hundreds of thousands of conversation logs exposed through internet searches last year.

A Google spokesperson told the BBC that the company does not control "which pages are made public on the web," noting that this decision rests with the websites themselves. He explained that Google gives site owners clear controls to decide whether their pages can be crawled and indexed, and that the company always respects those directives. Since the indexing of these Claude links is no longer happening, Anthropic has likely used the available tools to quickly block search engine access to those links, a process Google describes as straightforward but which must be initiated by the site owner itself. Other search engines where Claude chats also appeared, such as Bing, Brave and DuckDuckGo, were contacted by the BBC for comment.