Hitting a NERVE with attacks on AI-powered brain-computer interfaces (anil.recoil.org)

A recent study introduces a series of vulnerabilities termed "NERVE attacks" that could drastically impact the security of AI-powered brain-computer interfaces (BCIs). Led by Zahra Tarkhani and her team, the research outlines over 300 vulnerabilities across the BCI tech stack, underscoring significant risks as these technologies evolve. Notably, the study demonstrates various attack vectors such as Neuro-mimetic Forgery, which can generate counterfeit brain signals that bypass user verification, and Replay-based Hijacking, where recorded commands are illicitly replayed to manipulate devices without a user's knowledge. The implications are considerable, given that attacks could be performed rapidly and with minimal expertise, especially with the advent of tools like 'EEGle', which simplifies security assessments for BCIs. This research is especially timely, as it reveals alarming security gaps at a moment when BCIs are being adopted for critical applications like wheelchair navigation and robotic arm control. The demonstrated ease of executing multi-step attacks within milliseconds raises concerns over safety, particularly since many current defense mechanisms inadequately address these vulnerabilities. The study serves as a crucial call to arms for developers and engineers to prioritize the security of BCI systems, particularly as they gear up for applications in life-critical settings such as prosthetics and assistive technologies.