Original title: NERVE Attacks: Breaking AI-Powered Brain-Computer Interfaces
Zahra Tarkhani, Georgios Akkogiounoglou, Lorena Qendro, Isabel Tscherniak +1 · September 2026
Brain-Computer Interfaces, or BCIs, use artificial intelligence to connect a person's brain signals directly to physical devices. This setup creates dangerous, poorly understood security risks. Hackers could steal sensitive brain data, violate mental privacy, threaten physical safety, or hijack connected hardware. To study these threats, researchers introduced NERVE Attacks. This system organizes five distinct types of attacks across the entire BCI process: faking brain signals, desynchronizing systems to evade detection, replaying old signals, tapping data, and embedding hidden backdoors. The researchers also built an AI-assisted testing tool named EEGle. Testing revealed 17 new brain-specific attacks. They also found that generative AI makes it much easier for non-experts to launch these attacks against personal devices.
Imagine you're giving a friend directions over the phone. Normally you speak clearly and they follow along. Now suppose someone else could listen in on the call, pretend to be you, record your voice and replay it, make you say the wrong street, or secretly take over the call. In this paper, the phone call is the connection between a person's brain and a computer, and the five attacks are exactly those five ways of eavesdropping, impersonating, replaying, confusing, or hijacking that connection.
Brain-computer interfaces help people with paralysis move cursors or prosthetics, but the same neural signals that make them work can be stolen or manipulated. An attacker could read private thoughts or take control of a connected device, threatening mental privacy and physical safety. This matters for anyone whose brain data is collected, not just current BCI users.
The abstract does not specify whether the 17 attack instances were demonstrated on real BCI users or only in controlled or simulated settings. It also does not quantify how much generative AI lowers the barrier to entry, nor does it propose concrete defenses against the NERVE attack class. The stealth-effectiveness spectrum is described but not compared to existing security benchmarks.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
NERVE Attacks: Breaking AI-Powered Brain-Computer Interfaces — Zahra Tarkhani, Georgios Akkogiounoglou, Lorena Qendro et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.