Recent discoveries revealed that user conversations with Anthropic's AI chatbot, Claude, were accessible through public Google searches. This occurred because Claude's "share" feature, intended for sharing chats with specific individuals via a link, failed to prevent indexing by search engines. Hundreds of conversations, some containing sensitive information like legal admissions and crypto keys, were found across at least 25 pages of search results.

Redditors initially identified the vulnerability, noting that a simple Google search targeting the `site:claude.ai/share` directory exposed these chats. While Anthropic claimed shared content is publicly accessible and subject to archiving, the issue appears to stem from a lack of "noindex" tags on shared pages, a standard practice for preventing search engine indexing. This oversight, potentially indicating "gross incompetence," has raised significant privacy concerns.

Anthropic has since seemingly addressed the immediate issue, with previously exposed chats no longer appearing in searches. However, users are advised to avoid the share feature and remove past shared conversations. This incident starkly contrasts with Apple's privacy-focused approach in its AI offerings, like Apple Intelligence and Siri, which prioritize on-device processing and a secure "Private Cloud Compute" (PCC) system, ensuring no data is accessible or retained by Apple or Google.