Your shared Claude chats and Artifacts may be on Google
Hundreds surfaced via 'site:claude.ai/share'; some held medical records and wallet keys. Anthropic says it's fixed — but 'shared' never meant private.
Copy markdown
What surfaced, and how
Over the weekend, a plain `site:claude.ai/share` query in Google returned hundreds of publicly shared Claude conversations and Artifacts. The share pages went out without proper `noindex` tags, so anything a crawler could reach got indexed.
Chats or just Artifacts? Reports split
TechCrunch, Fortune and Gizmodo found full conversations; Axios says its Monday check surfaced only Artifacts — the apps, docs, spreadsheets and pages you build in Claude — not chat transcripts. Either way: if you shared it, treat it as public.
The exposed content wasn't harmless
Indexed pages included clinical-trial records with patient names, a document listing children's names and phone numbers, internal company files, employee reviews, source code, crypto wallet keys, and explicit material — the kind of thing you paste into a chatbot and forget.
Anthropic: not guessable, and now fixed
Anthropic says shared links aren't guessable or discoverable unless you post them somewhere public, that it doesn't hand sitemaps to search engines, and that the issue 'appears to have been fixed.' By Sunday the Google results had largely vanished.
You've seen this exact movie
It's nearly identical to the September 2025 ChatGPT leak that put roughly 100,000 chats on Google; xAI's Grok hit the same wall. The share button on every major assistant mints a public, crawlable URL — the failure mode is industry-wide, not Anthropic-specific.
Do this in the next five minutes
Open Settings → Privacy → Shared Chats, review every active share, and delete anything you didn't mean to publish. Going forward, assume a 'share link' is a public web page, not a private handoff.