Claude Share Links,
Exposed on Search Engines
"Only people with the URL can see this" — that's how most people treated Claude's share links, using them casually for work threads. But some of those pages turned out to be indexed by Google Search and viewable by anyone, according to multiple outlets. Here's what happened and what to do about it.
What was meant to stay private
ended up in search results
Claude's "Share" feature lets you hand someone a single URL to a conversation or an Artifact (a rendered, runnable piece of generated code or document). According to multiple outlets, some of these share pages were indexed by Google Search — meaning third parties who never received the URL could still find and open them with a single search operator, site:claude.ai/share.
It started on July 25, when a Reddit user ran that search operator and discovered a large number of strangers' shared chats showing up — reportedly more than 25 pages of results and over 200 conversations. TechCrunch and VentureBeat followed with verification pieces, and outlets including Fortune, Axios, and THE DECODER picked up the story. Reported exposed content included medical records, internal company documents, children's names and phone numbers, crypto wallet keys, resumes with real names, API keys, and a conversation with a lawyer asking whether they needed to self-report a violation.
| The assumption, until last week | What was actually happening |
|---|---|
| Only people with the URL can view it | Google crawled the page and listed the URL in search results |
| Casually used for work drafts and questions | Medical records, internal docs, personal data, and API keys became searchable |
| "Share" was assumed to mean limited access | Once posted to a forum or social media, anyone could trace the link |
| robots.txt was assumed to be enough | Excluding individual pages from indexing required a separate noindex tag |
The cause: one missing
meta tag
robots.txt controls site-wide crawl rules; the noindex tag stops indexing on a per-page basis. They are not the same thing.
Adding <meta name="robots" content="noindex"> to a share page's HTML tells search engines not to index it. The share pages in question shipped without that tag, so once a link landed somewhere crawlable — a forum post, a social media thread — Googlebot could follow it and index the full page, as THE DECODER reported. robots.txt governs whether a crawler may visit a page at all; noindex governs whether that page may be listed in results. Conflating the two is the technical root of this exposure.
An Anthropic spokesperson told TechCrunch that "shareable links are not guessable or discoverable unless people choose to share them themselves." Google's results for the affected pages began disappearing around July 26, suggesting a backend fix — likely adding the missing noindex tag. However, reports indicate some links were still surfacing on Bing and other search engines as of July 28, so the response wasn't uniform across search engines.
From discovery to containment
A Reddit user finds it
Running site:claude.ai/share turns up a large number of strangers' shared chats; they post about it.
Google results start disappearing
A likely noindex-tag fix takes effect; most affected pages drop out of Google search results.
Major outlets follow up
TechCrunch, VentureBeat, THE DECODER, Fortune, and Axios publish verification pieces; some links reportedly still appear on Bing.
How this hits your day-to-day work
The severity depends entirely on what you shared, and with whom. Here's what to check by role.
Engineers
Check whether design discussions or code reviews done in Claude — and Artifacts (runnable code) pasted into Slack or Notion — are still shared. Threads containing API keys or internal endpoints are the highest-risk exposure source.
Business & Legal
If you used share links to brainstorm contract drafts, HR matters, or customer data, confidential material may have briefly sat in search engines. Worth confirming with your security/legal team whether a report is warranted.
PMs & Product
If roadmap or unreleased-feature brainstorming was shared internally this way, non-public product strategy may have been externally visible. Worth revisiting your rules for keeping shares internal-only.
"Shareable links are not guessable or discoverable
unless people choose to share them themselves."
Why this isn't fully resolved
Concern 1: Response speed varies by search engine. Google's affected results began dropping around July 26, but reports indicate some links were still showing on Bing and other engines as of July 28. "Gone from Google" doesn't mean "fully fixed everywhere."
Concern 2: Deindexing isn't the same as revoking access. Removing a page from search results and revoking (unsharing) the link itself are two different actions. A link you shared in the past stays accessible to anyone who has the URL until you actively revoke it.
Concern 3: A gap between the "you control it" framing and reality. Anthropic frames share links as safe unless you personally spread them, but in practice, once a third party reposts a link to a forum or social media, a crawlable path exists regardless of the original user's intent. That gap between user expectation and actual technical behavior remains a design issue on Anthropic's side.
Concern 4: This may be an industry-wide weak spot. Around the same time, xAI's Grok was reported to have a similar chat-exposure issue, suggesting the assumption that "sharing means limited access" is a pitfall shared across multiple AI chatbot vendors, not unique to Claude.
That said, for casual personal brainstorming or general questions, the real-world risk is close to negligible. The severity scales almost directly with whether you created a share link containing sensitive information for work use. The single most reliable step is to audit your Claude account's share history now and revoke any link you're not sure about.