Key Points
- Claude chats with a public share link became web searchable as these settings removed indexing protections, an ongoing trend across AI services in the past year.
- The incident reflects a broader pattern in which confusing or permissive sharing features can expose private information, with AI tools increasing risk by correlating personal and enterprise data at scale.
- IANS Faculty recommend curbing shadow IT use of personal AI services, reviewing data sharing policies, and requiring vendors to provide stronger controls for AI services that touch an organization’s data.
Claude Share Feature Allowed Chats to Appear in Search Results
Users of Anthropic’s Claude may have had their chats with the LLM shared to the internet at large.
Commenters in a Reddit thread found hundreds of Claude chats publicly visible through Google and other search engines using straightforward search queries. These visible pages lacked "noindex" tags, enabling public propagation.
The chats were made public by users who clicked the option to share the link so that “anyone with the link could view” the chat. Anthropic has reportedly begun pulling the chats down from the public web, although some users and media outlets claimed the links are still working.
AI companies have struggled with public access to presumed private chats in the past year. Anthropic found itself under similar public scrutiny this time last year after it was discovered that user chats could be indexed and searched.
ChatGPT stripped a feature last year that published private chats to the web following user backlash. And xAI’s Grok exposed user chats last August before uploading entire user Git repositories this month, among other reported privacy violations.
Big Picture
Using publicly accessible internet services to share data inevitably creates exposure risk. The challenge for organizations is taking control through governance, technical controls and the elimination of shadow IT.
AI vendors are making headlines today, but this is a familiar problem -- other SaaS platforms in the past like Zoom, Trello, and Dropbox let users accidentally make private information public through lax or confusing sharing features.
"We must stop sharing data permanently using links that are meant to stay secret. If it's not a search engine, it will be a browser extension leaking it to something else. Only share with such public links when the data is not very critical and unshare as soon as possible.” Guillaume Ross, IANS Faculty.
The difference today between public exposure in years past comes from the speed and capability of AI services to interconnect data to users and ultimately their organizations.
Many organizations already have acceptable use or social media policies in place, but AI tools are increasingly blurring the line between personal and corporate activity. Data shared through consumer AI services can reveal insights about employees, projects, customers, and business operations outside established enterprise controls.
"With LLMs, the ability to instantaneously correlate everything anyone has ever posted online is now a commodity. Train users to understand there is no longer a clear delineation between what they post on their personal accounts versus corporate. They should take responsibility for their activity everywhere and always." Aaron Turner, IANS Faculty.
IANS Faculty Recommendations
- Update organizational policies on sharing with AI: Conduct periodic reviews of active shares for users who have enterprise access. Use DLP or CASB rules and tools where technically feasible to detect sensitive content flowing to AI domains.
- Unmask AI shadow IT: Discover where Claude (and ChatGPT, Gemini, etc.) is already in use, especially personal accounts processing company data. Prefer enterprise/tenant-controlled versions with SSO, audit logs, data residency options, and admin controls over share features. Disable or tightly restrict public sharing where the vendor allows it.
- Make accountable standards for AI vendors on data: Demand clear statements from AI providers on indexing controls, link expiration, authentication gates, and server-side revocation. Prefer architectures with private instances, VPC, zero-retention options, and enterprise plans where sharing is either disabled by default or gated behind organizational controls.
Authors & Contributors
Tim McCarthy, Author - Security Reporter, IANS News
Although reasonable efforts will be made to ensure the completeness and accuracy of the information contained in our News & blog posts, no liability can be accepted by IANS or our Faculty members for the results of any actions taken by individuals or firms in connection with such information, opinions, or advice.