Google’s search engine indexed dozens of Claude AI conversations over the weekend after a Reddit user highlighted a simple "site:claude.ai/share" query. The results revealed a range of sensitive material, from a detailed medical report bearing a patient’s name to a list of primary‑school children’s phone numbers, crypto‑wallet private keys and a lawyer’s question about professional‑conduct reporting.
Claude chats are private by default. Users can choose to share a snapshot of a conversation via a unique web address intended for a colleague or a small group. Anthropic, the company behind Claude, says that when a user clicks “share,” the content becomes publicly accessible, and the company does not provide search‑engine sitemaps or directories for those pages.
The problem, according to experts, is that Anthropic omitted a "noindex" tag that would tell crawlers to skip the pages. Without that directive, Google’s bots indexed the shared URLs, making them searchable like any other public page. When asked about the breach, Anthropic spokesperson Amie Rotherham emphasized that the system behaved as intended: "We give people control over sharing their Claude conversations publicly… These shareable links are not guessable or discoverable unless people choose to share them themselves."
Critics point out that the same share mechanism used by Google Docs includes a no‑index instruction, preventing documents from appearing in search results. Google’s own spokesperson, Ned Adriance, confirmed that site owners control crawl and index settings and that Google respects them. Anthropic’s decision not to use such a tag leaves a gap between technical compliance and user expectations of privacy.
The incident is not isolated. In September, Anthropic removed roughly 600 Claude conversations after they were indexed, and earlier this year OpenAI faced a similar issue when nearly 100,000 shared ChatGPT threads appeared on Google. Elon Musk’s Grok AI also suffered a comparable exposure. Anthropic has dealt with other security concerns this year, including a sandbox breakout in Claude Cowork that could read Mac credentials and a Grok Build bug that uploaded entire code repositories.
By Monday afternoon, the "site:claude.ai/share" search returned no results, suggesting Anthropic added the missing no‑index tag or otherwise blocked crawlers. However, any links that were previously captured can still be accessed directly, and cached copies may linger in third‑party archives.
Anthropic advises users to review shared chats in the Settings menu—Privacy → Shared Chats—and revoke any links they no longer need. The broader lesson underscores the growing responsibility of AI platforms to manage how a simple share button can inadvertently expose highly sensitive information.
Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.