Claude AI-Shared Erotic Role-Play and Legal Advice Surfaced on Google After Privacy Blunder

Anthropic's Claude AI is facing widespread backlash after a technical flaw exposed users' shared conversations on major search engines. The leak left private chats — including confidential legal advice and intimate role-play — publicly accessible online, forcing the company to address the security lapse.

Private Chats Surface in Search Results

Over the weekend, users were caught off guard after discovering that private Anthropic Claude chats were appearing in basic web searches. First spotted by a Reddit user, the issue exposed personal conversations covering everything from political advice and Kansas attorney ethics rules to intimate role-play sessions.

Through Claude's sharing feature, users can generate unique links to capture and share snapshots of their chat threads. Those URLs began appearing in search engine results because of the way standard web crawlers index publicly accessible links, combined with the unique ways people interact with generative AI.

Why Legal Advice and Role-Play Became Public

Search engines such as Google and Bing rely on automated web crawlers, which Anthropic instructs to bypass shared user chats. To do this, the company uses a standard robots.txt file — the traditional tool for marking specific areas of a website off-limits to web crawlers. Wayback Machine archives show Anthropic's file has explicitly blocked scrapers from indexing shared chat threads since at least September 2025.

Keeping specific pages out of search engine indexes, however, is far more complicated than many realise.

Searches for 'site:claude.ai/share' still return around 612 indexed pages on Bing, despite Microsoft's developer guidelines outlining a two-step blocking process. While Bing acknowledges that a robots.txt file discourages initial crawling, its official documentation states that site owners must also add a 'noindex' meta tag within individual page headers to guarantee pages remain excluded from search results.

Several conversations highlighted on Reddit were removed before WIRED could review them, while the specific search query that had returned Bing results had already stopped working on Google. Google's developer documentation explains that the search engine may ignore robots.txt rules when external websites link to a page, unless publishers also include a 'noindex' HTML tag or an 'x-robots-tag' in the page header.

An inspection of the leaked Claude chat pages by WIRED confirmed Anthropic had omitted the crucial noindex tag that both Google and Bing rely on to determine whether a page should appear in search results.

Google Points the Finger at Anthropic

Responding to the incident, Google spokesperson Ned Adriance placed responsibility on Anthropic, saying search engines simply process content that website owners make publicly available. He added that site administrators are provided with clear tools to control crawler access and stressed that Google honours those indexing settings when they are correctly implemented.

Anthropic declined to respond to questions about why it had failed to implement the 'noindex' tag across shared conversation pages.

Anthropic faced similar scrutiny last September, telling Forbes it relied on robots.txt to keep crawlers away from shared conversations. As the publication noted at the time, that standard directive does not guarantee individual pages will stay out of search engine indexes.

The Same Flaw Had Been Raised Before

Although robots.txt does not guarantee complete exclusion from search results, AI companies routinely rely on it for other protections. Developers are often told their online content will not be included in AI training datasets if they block specific crawlers through their site files — the same approach many of these technology companies use to protect their own platforms.