Anthropic let shared Claude conversations spill into Google search results, some carrying cryptocurrency wallet keys, names, and addresses. Reddit users found over the weekend that one search phrase returned long lists of shared chats and Artifacts, the mini apps built inside Claude, covering programming, work notes, and erotica, including one chat tagged “shared by Anthropic” in which Claude generated explicit content its own policy bars. Only chats run through the share button were exposed, since that button spins up a snapshot at its own URL, and a spokesperson told Fortune the links can’t be guessed and that sharing one makes the chat public like any other web page. The route is closed now, though many exposed links stay live. At this point it’s negligence. OpenAI, xAI, Meta, Google, and even Claude last year have already made the exact same mistake. It’s no longer an, “Oops! Who knew this could happen?” scenario. It’s now negligence.
