Users of Anthropic's AI chatbot, Claude, may have had their conversations exposed on Google over the weekend, a discovery made by Redditors who found a treasure trove of conversations between Claude users and their AI bots.
It appears that searching for a keyword followed by "site:claude.ai/share" revealed these conversations, although Anthropic quickly plugged the security hole, and searching for this term now returns no results.
However, this incident is not an isolated one, as it is the second time this year that Google has indexed Claude conversations, with a similar situation occurring in September.
The reason for this exposure is not entirely clear, but it seems that not all Claude chats had an equal risk of exposure, with Anthropic insisting that it does not share chat links directly with Google.
It appears that the vast majority of exposed chats were conversations that individuals had shared publicly on the open web, with Claude allowing users to share their chats if they wish.
However, this also means that private Claude chats that were not shared are unlikely to be involved in this latest exposure, and users who do not share their chats should be in the clear.
Further complicating matters, Claude's "artifacts" are still visible on Google, which are generated when the content of a project meets certain criteria and can include documents, code snippets, and other types of content.
Users who want to protect their Claude chats from future exposure can take action to ensure that none of their chats are indexed by a search engine in the future.
To do so, users should head to Settings > Privacy in Claude, choose "Manage" next to "Shared chats," and then select "Unshare" next to each chat they want to remove.
This will not delete the chats themselves, but will remove their public links and prevent them from being indexed by Google in the future.
