Anthropic Chatbot Conversations Exposed Online

Hundreds of user conversations with Anthropic's artificial intelligence chatbot Claude were found to be publicly accessible online. Links to these chats, which in some cases included personal and work-related information, appeared in search engine results from Google and other platforms.

The availability of the chat logs stemmed from users choosing to "share" a link to their conversations with Claude. These shared links were then indexed by search engines, making them discoverable. The issue was reportedly resolved over the weekend, but many of the accessible chats had already been saved and circulated online.

Anthropic stated that users retain control over whether to share their conversations. A company spokeswoman emphasized that shared links are not guessable or discoverable unless the user chooses to share them, and that shared content is publicly accessible and may be archived by third-party services. The interface for sharing a conversation indicates that "anyone with the link" can view its contents, but it does not explicitly state that the link may appear in search engine results.

Users on Reddit initially highlighted the issue, pointing to over 200 conversations with Claude that were discoverable through search engines. The content of these conversations spanned various topics, with users prompting the AI on subjects ranging from personal career advice to hypothetical scenarios.

Among the examples found were a user asking Claude about its allegiance to users versus the company, a request to draft an unpublished blog post containing corporate project details about cloud security, and queries about fictional transformations. Some users also included personal details like names, contact information, and work history when seeking assistance with resumes. Others appeared to use the chatbot for proprietary research, including transcripts of private conversations in fields such as healthcare.

This situation draws parallels to a similar incident involving OpenAI's ChatGPT last year, where chat logs also became publicly accessible. OpenAI subsequently adjusted its platform to limit the ease with which such logs could be discovered. A Google spokesperson clarified that the company does not control what pages are made public on the web, stating that websites owners have tools to manage whether their pages can be crawled or indexed, and that search engines respect these directives.

It is likely that Anthropic utilized available tools to block the chat log links from search engine indexing, a process that website owners can initiate through search engine platforms.