IMDEA Study Finds Anthropic Sends User Data to Third-Party Trackers Including Meta and TikTok

A privacy study by IMDEA Networks, titled "Prompt like a Butterfly, Sting like a Tracker" and accepted at PoPETs 2027, analyzed the web and mobile versions of nine AI chatbots in May 2026. The research identified that Claude transmits user data to third-party services, contradicting its positioning as a privacy-focused provider.

Data Leakage Findings

The analysis revealed specific privacy violations within Anthropic's web client:

• Third-Party Transmission: After accepting all cookies, the web client loads Segment Analytics via a first-party domain (a-cdn.anthropic.com) and forwards user events to eleven services, including Facebook, LinkedIn, TikTok, Reddit, and Google Enhanced Conversions. This server-side forwarding bypasses standard ad blockers.
• Cookie Rejection Failure: Claude continued connecting to Google Ads even after users explicitly rejected non-essential cookies. Additionally, chat links, email addresses, and other identifiers were sent to services like Datadog and Intercom, partly occurring even after cookie rejection.
• Paid Tier Limitations: Subscribing to a paid plan had minimal impact on web-based tracking. The only noted improvement was in the Android app, where paid users did not contact Intercom or Sentry, unlike free-tier users.

Scope and Context

The study examined ChatGPT, Claude, Gemini, Grok, DeepSeek, Perplexity, Le Chat, Meta AI, and Copilot. While Claude was not identified as the worst offender, and the paper did not show that conversation content itself was sent to ad networks, the transmission of metadata and identifiers to advertising platforms remains a significant concern. The study did not cover Windows or Mac desktop applications.