Some shared Claude conversations and Artifacts appeared in search results last weekend, turning a collaboration feature into a privacy cleanup job for users and IT teams.
This was not an account breach. The issue involved Claude’s share feature, which creates public links to chat snapshots and Artifacts that anyone with the URL can view.
Shared links became searchable
According to TechCrunch, Reddit users found shared Claude chats and Artifacts in Google Search by using queries such as “site.ai/share.” Some reportedly included health records, private company documents, and names and phone numbers of children.
Anthropic’s own sharing guide says Claude chats are private by default, but users can create shareable links. Once shared, anyone with the link can view the chat snapshot, including messages and Artifacts included before the link was created.
That wording can create a false sense of privacy. Many users treat “anyone with the link” as safe enough for a coworker, client, or friend. But if that link is later posted on a public forum, social media page, or other crawlable site, it can behave more like a public web page.
Artifacts make the exposure more serious because they can include dashboards, documents, apps, and project workspaces, not just chat text. Axios reported that some Claude-created apps, tools, and documents shared through public links appeared in Google Search.
The immediate lesson is simple: users should not assume an AI share link is private just because it was created for a small group.
What users and teams should check now
Claude users should open Settings, go to Privacy, and review both Shared Chats and Shared Artifacts. Anything containing names, phone numbers, health details, financial data, company documents, credentials, code, or personnel information should be unshared unless it was truly intended for public access.
Users should also search public posts, forums, team pages, and social accounts where they may have pasted Claude links. Unsharing a Claude link can stop future access to the snapshot, but it may not remove cached results, archived copies, screenshots, or links reposted elsewhere.
For companies, the risk is not limited to one employee’s chat history. Shared AI conversations can contain customer records, internal plans, source code, legal notes, or regulated data. A public AI share link can quickly become a data-handling problem, even when no account was hacked.
Security teams should tell employees to audit shared Claude links now and avoid using public share links for confidential work. For collaboration, employees should move outputs into systems with proper access controls, such as internal documents, wikis, ticketing tools, or approved project spaces.
The incident fits a broader enterprise AI control problem. eWeek recently reported that AI governance gaps are already leaving companies with weak visibility into AI systems and activity. As more employees use AI tools for real work, public sharing features need the same policy attention as file-sharing, email forwarding, and cloud storage permissions.
Organizations using AI agents or coding tools should also review access and identity controls. Neo’s launch showed how autonomous software is creating new questions around who owns an AI system, what it can access, and whether that access can be revoked.
Anthropic’s help page says Team and Enterprise users can share chats only with members of the same organization, not publicly. Admins should still confirm what controls exist, whether public sharing can be disabled, and how shared Artifacts are logged.
For now, the safest rule is simple: treat every AI share link as public unless your organization can prove otherwise.
Also read: ChatGPT share links were abused through real OpenAI URLs, showing how trusted AI platform links can create security risks even when the platform itself is not breached.