Claude chats exposed in Google search results via share feature

Over the weekend, members of a Claude discussion community on Reddit discovered that typing a specific search phrase into Google would return a long list of shared Claude conversations. The exposed chats included Claude's Artifacts—interactive tools and mini apps built within Claude—and covered topics ranging from erotica and programming to work notes and fake book reviews. Some conversations contained sensitive information including cryptocurrency wallet keys, personal names, and addresses. Notably, one chat labeled "shared by Anthropic" showed Claude generating explicit content, which violates Anthropic's stated policy on erotica.

Anthropić's share feature works similarly to competitors': when users select the share button, Claude creates a snapshot of the conversation at a unique web address, intended to be sent to one person or a small group with that link. The links do not grant access to a user's full account or all their conversations—only the specific shared chat. In response to Fortune, an Anthropic spokesperson explained: "We give people control over sharing their Claude conversations publicly, and in keeping with our privacy principles, we do not share chat directories or sitemaps with search engines like Google. These shareable links are not guessable or discoverable unless people choose to share them themselves."

Anthropic moved quickly to address the issue, and the links are no longer available via Google through the search technique shared on Reddit. However, at the time of reporting, many of the previously exposed chat links remained live and accessible to anyone who already possessed the direct link. The vulnerability is not new: the same type of problem affected Claude last year, and a nearly identical incident exposed almost 100,000 ChatGPT conversations on Google. Elon Musk's Grok chatbot has also been affected by the same issue in the past. The underlying cause in each case is that the "share" feature creates a unique URL, and these links were automatically published and left open to search engines, often without users' awareness. The repetition across OpenAI, Anthropic, and xAI suggests the industry has encountered persistent difficulty in permanently solving the problem.

The incident underscores a recurring tension in AI chatbot design: the "share" feature is intended for users to send a single conversation to a specific person or small group, but the technical implementation—a publicly accessible unique URL—creates an unintended surface for search engine indexing. Anthropic states it does not share chat directories or sitemaps with search engines and notes that shareable links are "not guessable or discoverable unless people choose to share them themselves." However, the fact that these links were indexed suggests that either search engine crawlers found and indexed them through other means, or that the default configuration inadvertently exposed them to indexing.

What makes this particularly significant is the pattern across vendors. OpenAI, Anthropic, and xAI (Elon Musk's company) have each encountered similar issues, indicating that this is not an isolated implementation mistake but a structural problem the industry has struggled to solve. Users increasingly treat chatbots as spaces to explore sensitive topics—legal questions, health concerns, work strategies—in a way they would not with a public document. The gap between that user expectation and the actual visibility of "shared" content creates real privacy risk. Although Anthropic appears to have fixed the immediate indexing vulnerability, the persistence of similar vulnerabilities across multiple companies suggests that a more fundamental redesign of how these features interact with search engines may be necessary.

No comments yet. Be the first to share your thoughts!