Overview
- A Reddit post that surfaced Tuesday showed a site-restricted Google search returning thousands of claude.ai/share links that led to private Claude conversations.
- Some of the exposed chats contained highly sensitive material such as crypto wallet keys, Social Security numbers, and a lawyer’s query about self-reporting misconduct.
- Anthropic and Google say the chats were only public when users created or posted share links and Anthropic points users to a settings panel to unshare previously shared chats.
- Google removed the site:claude.ai/share results soon after the reports while users said Bing and Brave briefly continued to show indexed links before those traces receded.
- This episode echoes prior indexing incidents and highlights how link-based sharing, missing per-chat crawler tags, and default settings can make private AI conversations discoverable and put users at risk.