Cambridge-led security team names five attack families that break AI-powered brain-computer interfaces
A team led by Zahra Tarkhani (Microsoft), Georgios Akkogiounoglou (KTH Royal Institute of Technology), Lorena Qendro (Nokia Bell Labs), Isabel Tscherniak (TU Munich), and senior author Anil Madhavapeddy (University of Cambridge) published NERVE Attacks: Breaking AI-Powered Brain-Computer Interfaces on arXiv on 8 September 2026. The paper introduces a taxonomy of five attack families that together span the complete BCI signal chain: Neuro-mimetic Forgery, Evasion via Desynchronization, Replay-based Hijacking, Vein Tapping, and Embedded Backdoors. It releases EEGle, an AI-assisted extensible framework for systematic BCI security analysis, and reports 17 novel neuro-specific attack instances plus a stealth-effectiveness spectrum for BCI backdoor design. The submission lands four days after Germany's outgoing federal data protection commissioner used the BfDI's annual Bonn conference on 4 September 2026 to press for tighter European handling of neurodata, and one day after four senior clinician-scientists published a Comment in Nature Neuroscience setting out four ethical priorities for implantable BCI research. The empirical attack surface has now caught up with the governance framing.