A Reddit user discovered on July 25 that a basic Google search could return hundreds of fully readable Claude conversations shared through Anthropic’s link-sharing feature.

The exposed chats included crypto wallet credentials, apparent Social Security numbers, legal discussions, and other sensitive information. The issue stemmed from a missing noindex meta tag. Anthropic’s robots.txt file was already in place, but it could not stop indexing once shared Claude links appeared on public websites.

Claude labels each shared conversation link with the phrase: "Anyone with the link." In practice, that also included Google. The distinction matters because an unlisted link is not the same as access control: it depends on obscurity unless the service also blocks search indexing and gives users a way to revoke access.

The discovery came when a Reddit user searched Google for site:claude.ai/share, a query that asks Google to show pages it has indexed from that specific section of a website. The search returned hundreds of other users’ conversations. The Reddit thread, posted at https://www.reddit.com/r/ClaudeAI/comments/1v6fiyj/you_can_view_a_lot_of_shared_conversations_via/, reached 4,000 upvotes.

The indexed material became a public record of what some users had shared with an AI assistant. Examples included a lawyer asking whether they needed to self-report a professional ethics violation, a user pasting a crypto wallet seed phrase—the master credential that controls access to cryptocurrency funds—and a person in the middle of becoming a nine-tailed fox. The results also included incestuous erotica and many other conversations.

One example quoted in the exposed results asked: "Why is the record industry hiding the fact that birds are secretly professional rappers who co-wrote most of Jay-Z albums?" Other examples ranged from users asking AI to help create slop to people arguing with the AI over politics.

The technical cause was a missing line of code. Anthropic’s share feature was intended to function like an unlisted YouTube video: the sender shares a link, the recipient can read it, and strangers should not be able to find it through search. The search-exclusion part requires a noindex tag, a single HTML instruction telling search engines not to list a page. Anthropic had not added that tag, so any Claude share link that appeared on a public forum, a tweet, or a Slack workspace could become an indexed webpage.

Anthropic’s robots.txt file, the standard file websites use to tell search crawlers which pages to avoid, was already blocking Claude share URLs. That measure was not sufficient.

Google’s documentation explains the gap: blocking crawling does not necessarily prevent indexing if a URL appears elsewhere on the web. If a crawler cannot open the page, it also cannot see a noindex instruction inside that page. As a result, some Claude conversations appeared in Google search results with the label "No information is available for this page." Google knew the link existed from third-party posts, but it could not display page details.

The same issue also applied to Claude’s published artifacts, including apps, dashboards, and financial tools built 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. Those artifacts show how AI sharing tools can expose not only prompts and replies, but also files, prototypes, and business data users build around an assistant.

A similar incident affected ChatGPT 11 months earlier. OpenAI’s "Make this chat discoverable" checkbox sent thousands of conversations to Google, Bing, and DuckDuckGo until OpenAI CISO Dane Stuckey removed the feature, describing it as "a short-lived experiment to help people discover useful conversations." Malwarebytes reported on that incident at Many of those chats were captured by the Internet Archive, a nonprofit that permanently stores copies of webpages, before OpenAI could remove them and remain accessible today.

Anthropic had already faced privacy-related scrutiny in 2026. In April, the company asked users for government IDs, as previously reported by Decrypt at https://decrypt.co/364509/claude-anthropic-government-id-kyc-privacy. In July, it quietly removed hidden tracking code from Claude Code after researchers found it, according to https://decrypt.co/372977/anthropic-removes-hidden-claude-code-tracker-privacy.

Google began removing Claude results by July 26 after Anthropic updated its configuration and added noindex tags. However, Bing was still surfacing shared links at the time of publication. Anthropic had not issued a public statement by publication time.

The fix does not erase copies that were already saved. A GitHub repository called Shared-Claude-Chats, available at https://github.com/fuadmefleh/Shared-Claude-Chats, had already archived 453 Claude conversations and 519 Grok chats from the exposed links, totaling 11,241 messages in plain text.

Users who want to revoke their own shared Claude chats can go to Settings → Privacy → Shared Chats. That stops future access, but it does not delete anything already saved elsewhere.