Large Language Models; Adversarial Attack; Red Teaming; LLM Hallucination; Robustness Evaluation; Trustworthy AI - NeurIPS 2025

today local_bar

Large Language Models; Adversarial Attack; Red Teaming; LLM Hallucination; Robustness Evaluation; Trustworthy AI

1 paper across 1 session

Poster Session 6

Friday, December 5, 2025 · 4:30 PM → 7:30 PM

Exhibit Hall C,D,E

SECA: Semantically Equivalent and Coherent Attacks for Eliciting LLM Hallucinations

#1305 · Buyun Liang, Liangzu Peng, Jinqi Luo, Darshan Thaker, Kwan Ho Ryan Chan, Rene Vidal

We propose SECA, a constraint-preserving zeroth-order method that elicits LLM hallucinations via semantically equivalent and coherent rephrasings.