Anthropic left Claude chats open to Google — crypto keys, legal queries exposed

For a brief window this week, anyone with a search engine and the right query could read private Claude conversations. A simple Google search for site:claude.ai/share turned up thousands of shared chats and user-created artifacts — documents, apps, and other content — that Anthropic’s Claude chatbot had generated. The culprit is an old, almost embarrassing mistake: the “Share with link” feature didn’t include a noindex tag, the tiny bit of code that tells crawlers to move along.

This isn’t a novel failure. OpenAI made exactly the same error last year with ChatGPT shared links, eventually pulling the feature entirely before reintroducing it with proper guardrails. That Anthropic shipped without this basic protection is the kind of oversight that makes you wonder what else got missed in the rush to ship.

Some of the exposed material was genuinely sensitive. Users on Reddit reported finding chats containing cryptocurrency private keys and detailed legal queries — the sort of thing you absolutely do not want indexed by Google. Anthropic moved fast once the issue surfaced. The results vanished from Google search quickly, though Bing and Brave Search held onto them longer, a reminder that scrubbing something from the internet is never instant or complete.

The incident also highlights how users treat AI chat interfaces as private workspaces, despite sharing features being fundamentally public. Anthropic points users to Settings → Privacy → Shared Chats to review and manage what’s out there. But the responsibility shouldn’t fall entirely on the person clicking “share” when the platform fails to provide the bare minimum of web hygiene. If you’ve used Claude’s share feature, now’s a good time to audit what’s floating around — and maybe reconsider what you paste into any chatbot.