Study identifies universal adversarial triggers that degrade LLM utility via endogenous preferences
A new arXiv paper proposes using endogenous preferences as vulnerability signals to discover universal adversarial triggers that degrade large language model utility when untrusted text enters prompt context.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-17 05:00 · arXiv cs.CL
Endogenous Preferences as Vulnerability Signals: Discovering Universal Adversarial Triggers for LLM Utility Degradation