SoylentNews

cereal_burpist [soylentnews.org] writes:

A missing line of code meant "share with a link" and "searchable by anyone" were the same thing—and someone already saved 11,241 of those messages to GitHub.

• A Reddit user discovered on July 25 that a simple Google search returned hundreds of fully readable Claude conversations.
• The chats exposed crypto wallet credentials, apparent Social Security numbers, legal discussions, and other sensitive data shared via Anthropic's link-sharing feature.
• The root cause was a missing noindex meta tag; Anthropic's existing robots.txt file was in place but couldn't prevent indexing once share links appeared on public sites.

A Reddit user ran a simple search—site:claude.ai/share—which asks Google to show everything it has indexed from a specific website—and got back hundreds of other people's conversations.

The thread [reddit.com] hit 4,000 upvotes and briefly became a public record of the things people apparently feel comfortable saying to an AI: a lawyer asking whether they needed to self-report a professional ethics violation, someone pasting a crypto wallet seed phrase, a person mid-process of becoming a nine-tailed fox, and enough incestuous erotica to surprise people who thought they'd seen the internet.
[...]
The cause was one missing line of code. Anthropic's share feature works like an unlisted YouTube video—you send the link, the recipient reads it, strangers can't find it through search. The "not findable through search" part requires a noindex tag, a single HTML instruction that tells search engines not to list a page. Anthropic hadn't added one, so every share link that landed on a public forum, a tweet, or a Slack workspace became a fully indexed webpage.

Anthropic's robots.txt file—the standard text file websites use to tell search crawlers which pages to skip—was already blocking Claude's share URLs. That sounds like it should have been enough but it wasn't.

Google's own documentation explains why: blocking a crawl doesn't prevent indexing if the URL appears elsewhere on the web. The crawler can't open the page, which means it also can't see the noindex instruction sitting inside it. Conversations appeared in search results labeled "No information is available for this page"—Google knew the link existed from third-party posts, it just couldn't describe what was in it.

The same gap applied to Claude's published artifacts—the apps, dashboards, and financial tools people build inside Claude. A search for site:claude.ai/public/artifacts surfaced payroll spreadsheets with employee names, internal customer relationship management chat exports, and unreleased product roadmaps.
[...]
Google began pulling Claude results by July 26 after Anthropic updated its configuration and added noindex tags, which sadly means you can’t snoop around other people’s conversations using this trick anymore… on Google.

However Bing is still surfacing shared links. Anthropic had not issued a public statement by time of publication.

A GitHub repository [github.com] ...called Shared-Claude-Chats has already archived 453 Claude conversations and 519 Grok chats from the exposed links: 11,241 messages, all in plain text.

Original Submission