Overview
- Published measurements on Oct. 13 showed six of nine web chatbots and multiple Android apps transmitted conversation-related metadata or identifiers to external analytics, advertising or tracking services.
- Researchers documented what was sent: auto‑generated chat titles, persistent chat links or IDs, user identifiers such as user IDs or emails, and in some cases prompts or screenshots that can reveal sensitive topics.
- The studies named specific services involved, including ChatGPT, Claude, Gemini, Grok and Perplexity, and identified third‑party recipients such as Google Analytics, Datadog and Microsoft Clarity.
- Built‑in privacy controls gave limited protection in the tests: rejecting nonessential cookies left roughly 80.8% of observed trackers active and private chat modes reduced but did not reliably stop data flows.
- Teams have notified European regulators, Spain’s AEPD urged EDPB discussion, vendors have offered mixed responses, and the findings will inform ongoing regulatory review and technical fixes.