Study Shows Multi-Agent Debate Can Poison LLM Fact-Checking
A new arXiv paper describes a game-theoretic multi-agent debate method that can deceive large language models' fact-checking judgments, highlighting risks from knowledge attacks on AI reliability.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.AI
Debate to Deceive: Poisoning LLM's Fact-Checking via Game-Theoretic Multi-Agent Debate