Debate-to-Deceive Framework for Fact-Checking Attacks
November 9, 2026
The Debate-to-Deceive (D2D) framework uses multi-agent evolutionary games to generate semantically coherent and rhetorically effective adversarial passages to bypass LLM fact-checkers.
HOW THIS AFFECTS YOU
●
researcherThis introduces a new multi-agent method for testing model robustness against rhetorical attacks.
●
policyYou need to consider how multi-agent debate can be used to bypass safety and fact-checking layers.