Intent Drift in LLM-Assisted Brain Computer Interface Communication: An In-Silico Benchmark Under Simulated Decoder Corruption
Language-model post-editing produced fluent semantic substitutions that rose with corruption, confidence did not reliably flag, and no interface policy removed, and this does not demonstrate clinical harm; prospective human-in-the-loop evaluation is needed.
A. Gorenshtein, M. Omar, E. Jia et al.
· medRxiv · 0 citations