AIToday
Large Language ModelsAI Safety & AlignmentITmedia AI+Published: Nov 17, 2026, 13:01 JST

42 AI chat services tested, 21 leak chat titles to trackers

Send on LINE
42 AI chat services tested, 21 leak chat titles to trackers

3 Key Points

  1. What happened

    Researchers examined 42 AI chat services including ChatGPT, Claude and Gemini; 21 leaked auto-generated conversation titles to external trackers, and in 9 the raw prompt was sent out too.

  2. Why it matters

    The leakage comes from ordinary page-title collection, not hacking, so users who confide sensitive questions may be sharing them with ad and analytics firms without knowing.

WHO IT HITSPeople who use AI chatbots for health, legal or personal advice may have their question topics exposed to ad and analytics companies. Privacy and security teams at companies deploying these chat tools may need to reassess what is embedded in the pages.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The study, accepted at IEEE S&P 2027, looked at 25 general-purpose AI services and 17 specialized health, legal and education services. Rather than only logging outbound traffic, the researchers traced browser JavaScript execution to identify which user action triggered a leak, which page element the data came from, and where it was sent.

The mechanism is mundane: trackers rely on the standard web behavior of reading a page's title tag. Because many chatbots summarize each question into a conversation title and display it as the page title, trackers mechanically read and transmit that title when a prompt is submitted or a chat is switched. For raw prompts, the paths were screen-recording tools that collect on-page text and prompts embedded in URLs.

On the general-purpose services, an average of 540 browser communications per service were observed, of which 118, about 22%, were tracker-bound. The concentration at Google, Microsoft, ByteDance and Meta suggests the exposure flows through widely used advertising and analytics infrastructure rather than any single chatbot vendor's own systems.

FAQ
Which companies received the leaked data?
Across the 25 general-purpose AI services, 41.8% of tracker communications went to Google, followed by Microsoft (12.2%), ByteDance (8.6%) and Meta (8.3%).
How does the leak happen?
Trackers use the standard web function that reads a page's title. Many chatbots turn the user's question into a conversation title, so the tracker reads and sends that title externally when the prompt is submitted or the user switches chats.
Were email addresses also exposed?
Yes. Among services where conversation leakage was confirmed, 11 of 13 general-purpose AI services and 8 of 8 specialized AI services also sent the user's registered email address externally.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleAgile Robots raises $500 million at $8.5 billion valuation