This research introduces PMIYC (Persuade Me If You Can), a novel framework for evaluating both persuasive capabilities and susceptibility in LLMs through multi-agent interactions.

• LLMs demonstrate persuasive abilities that rival human-level persuasion
• The framework uses Persuader and Target agents in multi-turn conversations
• Experiments reveal concerning vulnerabilities to misinformation in current models
• Results highlight the need for better safeguards against harmful persuasion techniques

Security Implications: This work exposes critical security concerns about how LLMs can be manipulated through persuasion, potentially compromising their alignment with ethical principles and creating vectors for misuse in real-world applications.

Persuade Me if You Can: A Framework for Evaluating Persuasion Effectiveness and Susceptibility Among Large Language Models