Researchers build a machine learning detector for Internet routing failures using only publicly visible data

Engineers built a machine learning system that can spot when Internet routing unexpectedly changes, using only data a traceroute command produces — no special access needed. This matters because Internet operators currently have to rely on expensive control plane monitoring tools that only they can access; a method that works from the outside could let independent researchers, security teams, and smaller ISPs spot routing problems themselves.

Internet routing instability — when packets suddenly start taking different paths — is hard to detect from the outside because operators keep detailed routing information private. This approach solves that by training an ensemble model to recognize the signature of route changes in latency data alone. That means the first publicly usable detector for routing failures could now exist, shifting the ability to monitor Internet health from exclusive operator access to anyone with command-line tools.