User Conversations with Claude AI Found in Google Search
In an unexpected revelation, users sharing discussions with an AI chatbot may unwittingly expose their conversations to broader viewership on the internet.
A significant repository of interactions with Anthropic’s Claude AI recently surfaced in Google’s search results, making them readily accessible for a fleeting period.
This past Saturday, a Reddit user reported that by entering “site:claude.ai/share” into Google’s search bar, one could uncover hundreds of Claude conversations.
The discourse ranged from outlandish conspiracy theories to requests for AI-generated adult content, often revealing the first name or username of each participant involved.
This discovery incited some users to criticize Anthropic for potentially neglecting user privacy. A parallel situation arose last year with OpenAI when user dialogues on ChatGPT, generated via public links, became indexed by Google. Following that incident, OpenAI removed the option to allow chats to surface in Google searches.
In contrast, Anthropic contends that the sharing functionality did not inherently submit link details to search engines like Google.
Instead, the company claims users may have disseminated links to shared conversations on public forums or social media, subsequently allowing Google’s crawlers to index them.
“We provide users control over the public sharing of their Claude conversations, and consistent with our privacy commitments, we do not supply chat directories or sitemaps to search engines such as Google,” reaffirmed Anthropic.
The Claude share feature is entirely optional, permitting the creation of public links to conversations that can be viewed by anyone. Anthropic attempts to prevent search engine crawlers via its Robots.txt file, which disallows crawling of claude.ai/share.
However, Google explains: “While Google won’t crawl or index the content blocked by a robots.txt file, we might still find and index a disallowed URL if it is linked from other places on the web.”
Anthropic elaborated: “These shared links are not discoverable or guessable unless individuals opt to share them. When a conversation is shared, it becomes publicly accessible, potentially archivable by third-party services, similar to other public web content.”
Google has indicated that the indexing likely resulted from users posting shared Claude chats on the open web.
“Neither Google nor any other search engine dictates what content is made public on the web; these pages have been indexed across various search engines,” Google informed PCMag.
“We grant site owners explicit controls over whether their pages can be crawled or indexed, and we always adhere to those guidelines.” Additionally, websites can utilize a “noindex meta tag” to prevent a page from appearing in Google search results.
Potential Scope of Exposure
The extent of this exposure appears to be relatively limited. One user attempted to archive the indexed Claude chats, retrieving only 519 conversations.
In the wake of the controversy, Google and other search engines seem to have ceased indexing the shared Claude dialogues.
Nevertheless, some have observed Google indexing Claude “artifacts,” or AI-generated documents/projects that are also shareable, although the users behind these creations may have intended to make them publicly visible.
For users concerned about their chats being indexed by Google, Anthropic offers the ability to disable shared chats and artifacts. To do so, navigate to Settings > Privacy > Your Data, where users can review which chats and artifacts have been shared and revoke access
If a Claude conversation is shared via a private messaging platform, Google’s web crawlers should remain unable to index it.
Source link: Au.pcmag.com.
