The Invisible Internet Project (I2P) is built to protect privacy, using garlic routing and a distributed network architecture to obscure who is communicating with whom. That makes it valuable for journalists, activists, and others with legitimate needs for anonymity. But the same features that shield users can also hide malicious activity, including the quiet theft of sensitive data.
A new paper on arXiv, "Detecting Data Exfiltration through I2P Anonymity Networks," takes on this dual-use problem. The authors propose a two-phase machine learning approach designed to identify data exfiltration traffic within I2P. The abstract frames the work as a response to the tension between I2P's privacy guarantees and the risk that those guarantees will be exploited.
The abstract does not include experimental details or performance metrics, so the effectiveness of the proposed method remains unclear from the announcement alone. Still, the paper highlights a growing area of research: building detection systems that respect the architecture of anonymity networks while trying to catch abuse. As I2P and similar tools gain users, such efforts will likely become more common.