Roughly 600 Claude AI conversations were just indexed by Google, exposed to anyone with a search engine, after users shared links publicly and Anthropic's crawl protections fell short.
What was exposed? Crypto wallet keys, API credentials, legal strategy discussions, internal company documents with staff names and emails, and deeply personal conversations. All discoverable with a simple Google search.
Anthropic says conversations are private by default and that indexing only happened because share links were posted publicly by users. They say they blocked crawlers and never submitted those directories to search engines.
But technical analysis suggests interactions between their robots.txt rules and noindex signals reduced the effectiveness of those protections. Bottom line, there's a big difference between "sharing a link with someone" and having it show up in Google results.
Results have largely been scrubbed as of today, suggesting Anthropic moved quickly. But removal from search doesn't mean the links are dead. Anyone who saved or bookmarked those URLs may still be able to access the conversations.
This isn't unique to Claude. Similar issues have hit ChatGPT and other AI chatbots. The lesson is, treat every AI conversation as potentially public.
Never paste private keys, passwords, or sensitive information into any AI chat, and if you use a share feature, assume anyone could eventually find it.
