Hundreds of private conversations with users of Anthropic’s popular artificial intelligence chatbot Claude were discovered to be publicly available online through simple Google searches, sparking fresh concerns about data privacy in the rapidly expanding AI chatbot industry.
Users on Reddit first discovered the breach over the weekend, finding that links to shared Claude conversations had been indexed by Google and other search engines. A search using the query “site:claude.ai/share” returned more than 200 conversations across at least 25 pages of search results, some dating from just weeks ago.
How the Exposure Occurred
Anthropic’s Claude chatbot includes a sharing feature that allows users to generate a link to a conversation, which they can send to colleagues, friends or family members. The feature tells users that “anyone with the link” can view the contents, but it does not explicitly warn that search engines may index and surface these links to the broader public.
Security researchers and WIRED, which reviewed a sample of the exposed pages, found that Anthropic had failed to include a “noindex” tag on shared chat pages – a standard web publishing control that tells search engines not to index a page. Both Google and Bing say they respect such directives when deciding whether to include pages in search results.
Sensitive Data at Risk
The exposed conversations contained a wide range of sensitive information. Some users shared personal details such as names, contact information and work history while seeking help with CVs and job applications. Others conducted what appeared to be proprietary research, including corporate project details and transcripts of private conversations.
In one conversation from April, a user prompted Claude to draft an unpublished blog post about cloud security that included details of a corporate project. Another user shared a lawyer’s question about whether they needed to self-report a breach of professional conduct. Some conversations even exposed cryptocurrency wallet creation details, including private keys.
When someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services.
Anthropic spokesperson
An Anthropic spokeswoman said the company maintains that users control when and if they share conversations. She said links are “not guessable or discoverable unless people choose to share them themselves.” The search indexing of the chat logs has since been removed, suggesting Anthropic used available webmaster tools to block the links from search results – a step it had not taken previously.
Industry-Wide Pattern
This incident mirrors similar privacy lapses by other major AI companies. OpenAI experienced an almost identical issue last year when ChatGPT chat logs were made publicly accessible through search engines, prompting the company to change how such logs were handled. Grok, the AI chatbot on X (formerly Twitter) owned by Elon Musk, also saw hundreds of thousands of chat logs exposed through online search last year.
A Google spokesman told the BBC that the company does not control “what pages are made public on the web,” adding that Google gives site owners clear controls to decide whether pages can be crawled or indexed, and always respects those directives.
What to Watch
The incident comes just days after reports that Claude’s “Cowork” feature could escape its macOS sandbox and gain full access to all files on a user’s system. Regulators in Europe and the United States are expected to scrutinize these back-to-back safety failures, potentially accelerating efforts to mandate privacy impact assessments for AI products before they reach consumers. Anthropic faces growing pressure to implement automated safeguards that prevent sensitive data from being inadvertently exposed through sharing features.