Proton's AI Paper Trail Tool Exposes Privacy Risks in AI Conversations
Proton has launched a free tool, AI Paper Trail, that analyzes exported ChatGPT and Claude conversations to reveal the extent of personal data inferable from user interactions.
Proton's new free tool, AI Paper Trail, aims to demystify the privacy implications of engaging with large language models (LLMs) by analyzing user conversation data. The tool processes exported chat logs from popular AI services like ChatGPT and Claude, generating a comprehensive privacy report that highlights what information can be inferred about individuals based on their interactions.
AI conversations, while seemingly innocuous on an individual prompt basis, can accumulate a significant amount of personal data over time. Proton's tool is designed to aggregate and interpret this data, revealing insights into users' work, relationships, habits, interests, purchasing behaviors, travel plans, and other sensitive details that might be inadvertently disclosed.
To demonstrate its capabilities, the tool was tested using personal ChatGPT conversation history. After exporting the data and uploading it to AI Paper Trail via a web browser, the tool analyzed the 200 most recent prompts. The analysis generated a personal privacy report, assigning an "AI Exposure Score" and estimating the advertising value of the revealed data.
In the test case, AI Paper Trail identified 47 distinct data points, assigning an exposure score of 58 out of 100, categorizing the user as "Leaving receipts." The report estimated the user's advertising value at $185 and flagged five specific "red flags" indicating areas of heightened data exposure.
The report's findings underscored how readily AI Paper Trail could piece together a detailed user profile from seemingly ordinary conversations. It identified data points across categories such as location, interests, technology usage, finances, and relationships, with location, interests, and technology use showing the most significant exposure.
While the tool makes inferences and its findings should not be treated as absolute facts, the exercise demonstrated the potential for seemingly unrelated pieces of information to coalesce into a comprehensive profile. Researching a topic or product does not automatically confirm a user's direct involvement, but the accumulation of such queries paints a detailed picture.
AI Paper Trail effectively translates an abstract privacy concern into a tangible, understandable format. By reconstructing details about travel destinations, purchasing habits, technology preferences, hobbies, and personal context from everyday questions, the tool offers a stark retrospective view of AI interactions and the digital footprint they leave behind.
Proton emphasizes that data uploaded for analysis is deleted immediately after processing and is not stored on their servers, addressing potential concerns about the tool itself becoming a data repository. This initiative aims to empower users with knowledge about the privacy costs associated with their AI engagements.