The route a shared link takes
The artefact in this case is a search pattern. TechCrunch reports that queries of the form site:claude.ai/share, circulating on Reddit, made Claude conversations and Artifacts that users had shared findable through Google. Among the visible material were a detailed medical report on a real patient, clinical trial results carrying patient names, documents with the names and phone numbers of primary school-age children, internal company files and employee reviews holding personal information. The timeline is short: flagged on Saturday, reported on Monday morning, and by Monday afternoon TechCrunch's own run of the same query returned nothing.[1]
Anthropic's statement makes two points: the company does not hand chat directories or sitemaps to search engines, and share links are not guessable unless people choose to share them. Both can be true and still leave out the mechanism behind this episode. The route TechCrunch describes is a link being crawled after it was posted elsewhere online. The two statements answer different questions: one is about who the company gave a directory to, the other about whether a page can be read by any crawler that reaches it.[1]
Who gave the consent?
In most of these examples the person who pressed share is not the person whose material became visible. The patient's medical report, the names in the clinical trial results, the children's phone numbers and the employee reviews were shared by third parties; those named never saw the button and never read a warning. The company's phrasing, which rests on people choosing to share links themselves, does not carry that distinction. The counter-reading: some of those sharing may have been authorised to publish the material, and the reporting does not say in which cases they were.[1]
The empty search on Monday afternoon is a fix, but it is worth naming what was fixed. A search result disappearing changes discoverability; it does not delete anything. If the link still resolves, everyone who copied it once still reaches the same page. Removal from an index is a findability measure, not the equivalent of taking the sharing back.[1]
What would close the gap?
What is missing here is a document rather than an accusation. A text describing the lifecycle of the share feature — what happens to the page when a link is revoked, what instruction is given to crawlers, how long caches and third-party copies persist — would close every question that cannot be answered today. If Anthropic publishes such a text by 31 October 2026, the next episode of this kind starts from a document rather than a claim; if it does not, each new example will leave the same gap behind the same two sentences.[1]