Nine of 42 AI chat sites send prompts to third parties
Original title: Penny for Your Prompts: A Measurement Study of Contextual Leakage in AI Chat Platforms
Twenty-one platforms leak conversation titles, and some also send emails that can identify users.
A measurement study audits the web interfaces of 42 general-purpose and domain-specific AI chat platforms. Browser instrumentation attributes each data flow to the triggering user action and the DOM capture method. Nine platforms leak verbatim user prompts to third-party trackers, and 21 leak conversation titles. Some platforms also send user emails, which can tie conversational context to an identity.
Unconfirmed: The abstract does not name the 42 platforms.
- Audited sites include health, legal, and education chatbots.
- Tracking covers analytics, advertising, and session-replay scripts.
- Attribution is done at the JavaScript engine, not only the network layer.
- User emails sent to trackers can link chats to identity.
Impact: Prompts, titles, and emails sent to third-party trackers
From the original article
1 min read at arxiv.orgAI chatbot platforms have become one of the most popular web interfaces users interact with online, routinely receiving personal and intimate information shared under an expectation of privacy.
By Paschalis Bekos · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.