Anthropic’s Claude Shared Chats and Artifacts Indexed by Google and Bing, Exposing API Keys and Health Records

A Reddit user flagged the problem over the weekend, demonstrating that a simple search operator like site:claude.ai/share on Google returned hundreds of shared Claude conversations. The discovery quickly spread across the AI community as developers and privacy researchers confirmed the results were not isolated to a single search engine. According to MLQ, the chats also appeared on Bing and Brave Search. Wired reports that the issue appeared to have been first flagged by a Redditor who realized these ostensibly private conversations were accessible without needing the direct link from the original owner.

What made the breach alarming was the sensitivity of the exposed material. TechCrunch confirmed that some chats contained health records, private company documents, and the names and phone numbers of children. Forbes noted that hundreds of Anthropic chatbot transcripts showed up in search results. The exposure wasn't limited to text either. The Decoder updated its reporting to confirm that user-created Artifacts, including documents and interactive mini apps built inside Claude, were also indexed and discoverable through search engines.

Reports from multiple outlets detailed the range of sensitive information that became publicly searchable. IBTimes documented exposed API keys, cryptocurrency wallet details, and CVs among the search results. Tech.yahoo confirmed that developers found crypto wallet data and access keys in the indexed chats, prompting some users to scramble to revoke old share links. The Decoder added that some conversations contained legal questions and crypto keys.

Cybersecuritynews reported that users could access conversations containing legal advice, engineering work, and personal discussions. The exposure was not a breach of private chats in the traditional sense. ZDNET emphasized that private chats were not leaked and that Anthropic reacted quickly. The mechanism was more mundane: the share feature generated URLs that were supposed to be unlisted but were missing the noindex meta tag that tells search engines to ignore a page. Without that tag, crawlers from Google and Bing treated the shared URLs like any other public web page and added them to their indexes.

Anthropic moved quickly once the exposure became public. ZDNET reported that Anthropic reacted swiftly to the situation. The fix was technically straightforward. The company added noindex tags to the shared chat pages, which instructs search engines to remove them from their indexes. Wired noted that the screwup shows how tricky it can be to stop web crawlers from making ostensibly private conversations with AI chatbots entirely too public.

By Sunday, most of the indexed results had been removed from Google, though the cleanup process took some time across different search engines. The Decoder noted that user-created Artifacts were exposed alongside the chats and that a Google search for site:claude.ai SEARCH TERM would still turn up many of these artifacts on Claude.ai even after the initial fix. The incident underscored a persistent gap in how AI companies think about share features. They treat a shared URL as private because it is unlisted, but the web treats any URL without a noindex tag as fair game for crawling.

For everyday users, the exposure is a blunt reminder that sharing a conversation with Claude creates a publicly accessible URL, not a private room. ZDNET advised users to be cautious about creating share links. The risk is not that someone guesses your URL, but that search engines will crawl and index it, making it discoverable by anyone with the right search query. TechCrunch described the share chat feature as a tool that allows users to create links that enable anyone with the assigned URL to view a conversation or project. The missing noindex tag turned that anyone with the URL into anyone with a search engine.

Users who shared sensitive information through the feature, including financial data, proprietary code, or personal health details, should assume those conversations were visible to anyone who found them before the fix. MLP reported that the verified exposure concerned chats whose users had activated Claude's share feature, meaning the exposure was opt-in but poorly understood. The practical takeaway is to revoke any old share links and treat the share feature as a public broadcast tool, not a private collaboration channel.

This is not a problem unique to Claude. Wired framed the incident as a demonstration of how tricky it is to stop web crawlers from making ostensibly private conversations with AI chatbots entirely too public. The underlying issue is a design philosophy mismatch. AI companies build share features with the mental model of an unlisted YouTube video, visible only to those with the link. Search engines operate on a different model: if a page exists and is not explicitly blocked, it is fair game for indexing.

Forbes contextualized the incident within a pattern of AI privacy lapses, noting that hundreds of transcripts showed up in search results. IBTimes reported that the discovery triggered debate across the AI community after developers demonstrated that some shared Claude pages could be found through search engines. The incident raises questions about whether AI companies are doing enough to protect user data from accidental exposure through basic web infrastructure, and whether features that feel private in the product are actually private on the open web.

Anthropic has not issued a formal public statement beyond the technical fix, according to Tech.yahoo, which noted the company had not addressed the matter as of their writing. The immediate risk appears contained with the noindex tags in place, but the long-term question is whether AI companies will redesign their share features with privacy as the default rather than an afterthought. The incident may also attract regulatory attention, particularly in jurisdictions with strict data protection laws.

For users, the cleanup is ongoing. Anyone who shared a Claude conversation should review their share history and delete links containing sensitive information. Search engines will eventually drop the indexed pages now that noindex tags are present, but cached versions may persist for some time. The episode leaves a lasting lesson: in the age of AI chatbots, sharing a conversation is publishing it, and the only true privacy setting is not to share at all.