What did the LeakyLM study discover about AI chatbots?

Researchers at IMDEA Networks Institute said major AI chatbots transmitted user-related data to third-party advertising and analytics services. The study identified more than 13 embedded trackers across platforms including ChatGPT, Claude, Grok and Perplexity.

Which AI chatbots were included in the study?

The research analysed ChatGPT, Claude, Grok and Perplexity. Researchers said all four platforms transmitted some level of user-related metadata to third parties.

What kind of user data was reportedly shared?

The study said some platforms transmitted conversation URLs, cookies, browser identifiers and metadata linked to chats. In some cases, trackers allegedly received preview information tied to conversation content.

Why are conversation URLs considered sensitive?

Researchers argued that leaked conversation URLs could effectively expose entire chats if those links remain publicly accessible. Anyone with access to a public conversation link may potentially read the contents without authentication.

Why was Grok described as the most exposed platform?

The report claimed Grok guest conversations were publicly accessible by default. Researchers also alleged that TikTok trackers received Open Graph metadata containing previews or screenshots linked to conversations.

Did the study prove Meta or Google actually read user chats?

No. The researchers said they did not confirm whether companies such as Meta or Google actively accessed or reviewed conversations. However, they argued that the technical infrastructure allowing potential access already exists.

Does rejecting cookies fully protect AI chatbot privacy?

According to the study, not always. Researchers claimed some tracking systems continued operating even after users rejected optional cookies or privacy prompts.

Why is this important for AI users?

Many users assume AI conversations remain private or isolated from advertising systems. The study raises concerns that sensitive prompts, personal information or business discussions could potentially become accessible through tracking infrastructure.

AI chatbots accused of leaking user data to ad trackers

Written by Heidi Cuthbert Published May. 8 2026

Researchers at IMDEA Networks Institute said several leading AI chatbot platforms were transmitting user-related data to third-party advertising and analytics services linked to companies including Meta, Google and TikTok.

The study, called “LeakyLM”, analysed major AI assistants including ChatGPT, Claude, Grok and Perplexity, identifying more than 13 embedded tracking systems across the platforms.

According to the researchers, some chatbot services transmitted conversation URLs, cookies and other metadata to third-party systems even when users rejected optional cookies or privacy prompts.

The report described Grok as the most exposed platform because guest conversations were reportedly public by default and TikTok trackers allegedly received Open Graph metadata containing previews or screenshots of chat content.

The researchers said ChatGPT and Claude used stronger access controls than Grok but still transmitted certain identifiers and conversation-related metadata to advertising and analytics providers.

Claude reportedly routed some advertising-related data through Anthropic’s own infrastructure rather than directly through browser trackers, potentially bypassing standard ad-blocking protections.

The study said it had not confirmed whether companies such as Meta or Google actively read private conversations, but argued that the technical infrastructure enabling access to the data already existed.

The findings have reportedly been submitted to data protection authorities while researchers said future analysis would examine additional AI platforms including Meta AI, Microsoft Copilot and Google Gemini.