Claude conversations containing cryptocurrency wallet details, API keys, CVs, and company documents appeared in Google Search results over the weekend, prompting initial claims of a major Anthropic data leak. The actual cause is more mundane but arguably more instructive for teams deploying agents in enterprise environments.
What Happened
Users on Reddit discovered that searching Google with site:claude.ai/share returned pages of Claude conversations. Screenshots showed sensitive data visible in search results. Developer Om Patel flagged the issue on X, writing: “When you use Claude’s Share feature, it creates a public link. It turns out those links got indexed by search engines, so ‘share with anyone who has the link’ actually became ‘anyone can find this by searching,’” according to India Today.
Claude private conversations remained private throughout. The issue affected only conversations that users intentionally shared via Claude’s public Share feature, which creates a publicly accessible webpage containing the conversation up to the point of sharing.
The Technical Gap
Anthropic’s robots.txt file blocks search engine crawling on shared chat pages, according to India Today’s investigation. The question is why some pages were indexed anyway. Possible causes include: third-party tools or browser extensions that bypass robots.txt restrictions, cached versions from before the directive was added, or external sites linking to shared conversations (which gives search engines a path to the URL regardless of crawl directives on the destination domain).
The Agent Data Governance Problem
The incident is minor in isolation. No private conversations leaked. Users who clicked “Share” created public content. But the pattern matters for agent deployments.
When employees use Claude or other AI agents in enterprise workflows, they generate conversations that may contain proprietary data, credentials, customer information, or internal strategy. If any of those conversations get shared, whether intentionally (“let me send this to my colleague”) or accidentally (wrong button click, Slack integration auto-sharing), that data becomes persistent and potentially discoverable.
The gap is not in Anthropic’s security. The gap is in user expectations. “Share with anyone who has the link” is a design pattern borrowed from Google Docs, where the content is understood to be a document. In an agent conversation, the content might be a full debugging session with production credentials pasted inline. The mental model that “sharing a link” equals “publishing to the internet” does not match how most users think about chat interfaces.
For enterprises deploying agents at scale, this points to a policy requirement: agent conversations containing sensitive data need the same data loss prevention (DLP) controls that email and cloud storage already have. The conversation is not just a chat log. It is a work product that may contain everything the agent processed.