A quiet but significant privacy lapse surfaced over the weekend when users discovered that shared Anthropic Claude conversations were appearing in Google and Bing search results. The issue — first flagged by a Reddit user — exposed chats covering sensitive topics: political affiliation advice, legal ethics questions, and erotic role-play.

How Shared Chats Work

Claude allows users to generate a public URL — a shareable "snapshot" of a specific conversation thread. The feature is opt-in, but the intention is that only people with the direct link can view the chat. The problem is that a public URL is still, technically, a public page.

The robots.txt Gap

Anthropic uses a robots.txt file to instruct web crawlers not to index shared chats. According to the Wayback Machine, this has been in place since at least September 2025. But robots.txt has a fundamental limitation that tripped Anthropic up here.

Both Google and Bing have documented that robots.txt alone is not sufficient to prevent a page from being indexed. Specifically:

  • Google states it may still index a page if it's linked to from elsewhere on the web — unless the page also includes a noindex HTML tag or an x-robots-tag in the HTTP response header.
  • Bing similarly recommends that developers include a noindex tag on individual pages in addition to robots.txt directives.

WIRED confirmed that the exposed Claude chat pages did not include a noindex tag — the missing piece that would have given search engines a clear page-level signal to suppress indexing.

Who's Responsible?

Google was direct in deflecting accountability.

"Neither Google nor any other search engine controls what pages are made public on the web, and these pages were indexed across many search engines. We give site owners clear controls to decide whether pages can be crawled or indexed, and we always respect those directives." — Ned Adriance, Google spokesperson

Microsoft (which owns Bing) and Anthropic did not provide comment. Notably, this isn't a first offense: Anthropic faced the same criticism in September 2024, responding to Forbes at the time by citing its robots.txt setup — even though the limitation of that approach was already well understood.

As of publication, Google results for the shared chat query had cleared up, but Bing still showed approximately 612 results via a site:claude.ai/share search. Critically, the pages WIRED reviewed still lacked noindex tags, meaning re-indexing remains a live risk.

What This Means for AI Builders and Founders

This incident highlights a critical — and underappreciated — gap in how AI product teams think about web infrastructure:

  • robots.txt is not a privacy control. It's a convention that well-behaved crawlers follow, not a security boundary. Treating it as one is a mistake.
  • Every shareable URL is a potential public page. If your product generates user-facing URLs for shared content, those pages need explicit noindex tags and/or authentication gates.
  • Search engines are unpredictable. Linked pages can be indexed regardless of crawl instructions. The only reliable server-side control is the noindex directive at the page level.

For AI startups shipping sharing or collaboration features, this is a sharp reminder that privacy defaults need to be enforced at the HTTP layer, not delegated to conventions that crawlers can override.

A Broader Pattern in AI Infrastructure

There's an ironic coda to this story. While Anthropic failed to block its own shared pages from being indexed, the company — like OpenAI and Meta — actively uses robots.txt to block competitors' AI training crawlers from accessing its site. Google's AI training crawler, Google-Extended, is explicitly disallowed by all three. Google declined to comment on that specific detail when asked by WIRED.

Claude users who want to audit their exposure can navigate to Settings > Privacy > Shared Chats to review and revoke any active shared links.