This research investigates how Large Language Models can be weaponized for persuasion through personalized arguments and fabricated data.

• Personalized persuasion significantly increases LLM effectiveness in changing human opinions
• Fabricated statistics further enhance persuasive capability, even when humans are aware of potential AI deception
• Security implications are substantial for disinformation campaigns and targeted manipulation at scale
• Ethical guardrails are urgently needed as these capabilities become more accessible

These findings highlight critical security vulnerabilities that organizations must address as LLMs become more widespread in consumer-facing applications and communication channels.

Tailored Truths: Optimizing LLM Persuasion with Personalization and Fabricated Statistics