Recently, an AI assistant named Claude under Anthropic was exposed to a incident where chat records were shared and indexed by search engines. A large number of user-generated conversations and Artifacts (interactive applications and documents created in Claude) appeared in Google search results, raising concerns about privacy and security of AI products.

The incident was first discovered by a Reddit user last Saturday. By using operators such as "site:claude.ai/share" on Google, users could retrieve some public sharing links of Claude. Some content involved sensitive information such as health records, internal corporate documents, children's names, and phone numbers. Subsequently, 404Media reported the issue on Monday morning.

According to the information, the problem is related to Claude's "Share Chat" feature. This feature allows users to create public links, and anyone who has the link can view the corresponding conversation or project. The Claude interface previously warned that "anyone with the link can view it," but this design mainly targets sharing with friends, colleagues, or small groups, not for public dissemination across the entire internet.

QQ20260728-090651.jpg

Anthropic responded that the share links would only appear in search results if users actively published them to forums, social media, or other locations accessible by search engines. The company would not proactively provide chat records or site maps to search engines like Google. An Anthropic spokesperson, Amy Rotherham, stated that after users chose to publicly share, the content could be indexed by third-party services like other public online content.

The exposed content includes code, work notes, company internal materials, medical reports, clinical trial documents, and employee reviews. Futurism reported that some public content even included real patient information and children's contact details. Additionally, Fortune reported that a chat record labeled "Anthropic Shared" showed that Claude generated explicit content, while Anthropic's usage policy explicitly prohibits generating explicit content.

By Monday afternoon, TechCrunch was no longer able to retrieve the previously public content when testing according to the relevant method, indicating that the issue may have been partially resolved. Google stated that the search engine cannot control which web pages are actively made public by websites. Website owners can control whether pages are allowed to be crawled and indexed through relevant mechanisms.

Claude, claude

Previously, similar AI chat privacy incidents have occurred. Last year, Forbes reported that hundreds of Claude chat records were indexed by search engines, and Google estimated that about 600 related conversations had entered the index; in the same year, 404Media also reported that researchers scraped about 100,000 publicly shared ChatGPT chats.

This incident once again highlights that as AI chat tools enter scenarios such as offices, healthcare, and education, the public sharing mechanism, data access management, and search engine indexing control will become important aspects of AI product security design. Users can also check the publicly shared chat links via "Settings → Privacy → Share Chat" in Claude.