ASRogue: Manipulating ASRank-Inferred AS Relationships

Accurate knowledge of Autonomous System (AS) business relationships is critical for many Internet security and networking tasks. Since these relationships are not publicly visible, operators and researchers rely on algorithms that infer them from empirical routing data. Yet, despite decades of widespread deployment, the security of AS relationship inference against adversarial manipulation remains largely underexplored. To close this gap, we conduct a systematic security analysis of ASRank, a foundational algorithm that underpins many established methods and widely used datasets. We show that ASRank's output is highly sensitive to small, targeted perturbations of AS triplets extracted from BGP paths. Carefully chosen triplet changes can alter inference order and trigger cascading misclassifications. Building on this insight, we present ASROGUE, a manipulation attack against ASRank-based inference. ASROGUE adaptively crafts AS triplets to steer inference order and outcomes, scales triplet injection via prefix splitting and AS-path poisoning, and embeds forged triplets into policy-compliant BGP announcements that can evade common routing defenses. Extensive evaluation shows that when the attacker controls at least two ASes, AS-ROGUE achieves an overall 96.7% manipulation success rate, exceeding 99% for small and medium-sized providers, and remains effective under constrained attack overhead and temporal variation. Finally, controlled experiments on the PEER-ING testbed validate the feasibility of launching ASROGUE attacks in the real-world Internet.