Claude User Chats Leaked Into Google Search

The security of AI services and user privacy have once again become the center of serious discussion. According to information distributed by 404 Media and social media users, it was discovered that an unknown portion of public chats in the Claude neural network and user-created Artifacts mini-apps were openly indexed in the Google search engine. This situation raised serious concerns regarding personal data security. This is reported by Techcrunch.com reports .

The issue turned out to be related to the service's "share chat" feature. This capability allows users to share their projects or chats with others via a special link. Although the interface includes a warning that anyone with the link can view it, many users did not imagine that this data would remain open to the entire internet and leak into search engines.

Disclosure of confidential data

Observations by TechCrunch and Futurism showed that among the pages indexed by the search engine were very sensitive and confidential contents. In particular, detailed medical reports of real patients, clinical trial results containing patient names, and names and phone numbers of primary school students were identified. It was also revealed that documents intended solely for internal company use became publicly visible.

Furthermore, among the exposed Artifacts documents, various programming codes, work notes, and AI-generated texts of an erotic nature were found. As Fortune reported, one such chat showed Claude generating explicit erotic content contrary to its own rules, which violates Anthropic's policy.

Companies' response and resolution of the problem

In a comment to TechCrunch, Anthropic representatives stated that such links only appear in search results if users themselves post them in places accessible to search engines, such as forums or social networks. According to company representative Ami Roterman, they do not share chat catalogs or sitemaps with search engines like Google, but if a user shares information publicly, it can be archived by third-party services.

Google representative Ned Adriance, in turn, emphasized that the search engine does not monitor the popularity of web pages, and search engines always follow instructions set by site owners. This glitch, discovered by a Reddit user on Saturday and widely covered on Monday, was promptly resolved. Test searches conducted starting Monday yielded no results, indicating that the issue was successfully fixed.