Some people's chats with Claude AI found publicly available online - BBC

What the report says

BBC News reported that hundreds of shared conversations with Anthropic’s Claude AI chatbot were found accessible through online search, including some chats containing personal, workplace or potentially sensitive information. The issue involved Claude conversations for which users had created share links; those links were then indexed by search engines and could be found using site-specific searches.

According to the BBC, users on Reddit first identified the exposed chats, which appeared across more than 200 conversations and at least 25 pages of search results. The logs reportedly included examples such as CV assistance containing names and contact details, work-related research, a draft blog post involving a corporate cloud security project, and other personal or unusual prompts. The search visibility was removed over the weekend, though the BBC said many links had already been saved and circulated online.

Anthropic told the BBC that Claude users decide whether to share conversations and said shared links are not discoverable unless users choose to make them available. The company’s spokeswoman said shared conversations become public content and may be archived by third-party services. The BBC noted that Claude’s share tool tells users that anyone with the link can view a conversation, but does not explicitly warn that the link may appear in search results.

The incident matters because shared AI chats can contain private, professional or proprietary material, and many users may not understand how public links can be indexed. The BBC also cited similar past cases involving ChatGPT and Grok. Google told the BBC it does not decide which web pages are public, but provides site owners with tools to control crawling and indexing.