Shared AI chats showed up on the open web. A chatbot is not a private notebook

Meta paused the sharing feature of its Muse assistant after a security firm found more than 6,800 Muse links by scanning public DNS records. The chats it reached included a marketing firm's revenue review, a small business loan application and a full-year staff calendar (The Verge).

The leak is not Meta's alone. A preprint by the IMDEA Networks Institute, not yet peer-reviewed and reported by netzwoche, examined nine chatbots: five passed chat permalinks to tracking services, often with email addresses or user IDs and even in incognito mode, and about 81 percent of the trackers stayed active although users rejected cookies.

The lesson: a consumer chatbot is not a private notebook. An employee who pastes customer data, contracts or financials into a free chatbot creates exactly this kind of leak. A business account with a data processing agreement, sharing links switched off and a short rule for staff solves most of it.

Job seekers paste their CVs, salary history and personal stories into chatbots every day. Knowing what to share and what not is a basic skill we teach everyone who comes to us after losing a job.

Practical step: write down which data must never go into a chatbot, switch off public sharing links in every tool, and move business use to accounts with a contract.