MirageBeam: Stealthy Adversarial Beamforming Feedback for Resilient WiFi Sensing Hijacking

Beamforming feedback information (BFI) sensing is emerging as a promising successor to channel state information (CSI)-based WiFi sensing thanks to its accessibility on commodity hardware without driver hacking. Yet, its security implications have been scarcely studied. Unlike CSI, BFI is broadcast in plaintext, offering adversaries a broader attack surface. We present MirageBeam, the first practical framework for adversar ial attacks on BFI-based sensing. By injecting carefully crafted BFI frames, adversaries can launch both untargeted and tar geted attacks that cause misclassification in behavior recognition systems. The core challenge is to maximize misclassification while remaining indistinguishable from regular behavior samples. We formulate an optimization problem that jointly selects the forged BFI amplitude and injection timing to covertly control the BFI sequence observed by the receiver. Furthermore, real world injections face the challenges of attack asynchrony caused by unpredictable injection delays. We introduce a delayed-BFI injection window and recast the optimization as a weighted formulation over delay units, enabling seamless adaptation to varying delays. Implemented on commodity WiFi hardware across three indoor settings, MirageBeam achieves over 81% attack success while maintaining a detection rate below 30%, revealing critical vulnerabilities in existing BFI sensing systems. Actionable countermeasures are ultimately outlined to safeguard future BFI sensing deployments.