🎙️ INSIDE AI Podcast: Can large language models be manipulated like a human?
Is an AI purely logical, or can it be "socially engineered" into breaking its own rules?
In this episode, we dive into groundbreaking research from Texas Tech University that explores the intersection of cybersecurity and human psychology.
We discuss how researchers used Robert Cialdini’s principles of influence —like reciprocity and social proof—to test if LLM models can be manipulated using the same psychological tricks that work on humans.
🕒 Episode Highlights:
- (00:32) - Can You Hack an AI with Psychology? Exploring whether persuasion and psychological tricks can make an AI agent do something it was programmed not to do.
- (01:19) - The Principles of Influence How foundational human influence tactics like reciprocity, authority, and scarcity apply to machine behavior.
- (02:06) - The "Free E-Book" Trap A startling example where the principle of reciprocity tricked an LLM into generating actual malware code in exchange for a "gift".
- (02:44) - The Three Stages of Deception From simple misleading answers to complex, context-aware "pro-social" manipulation.
- (04:18) - Building a Cognitive Firewall How Loop AI utilizes specialized language models and private data to defend against these psychological vectors.
🚀 Explore
Loop.AI:
- Join our team: Explore open positions at
Loop.AI/careers
- Build AI agents powered by custom Small Language Models (SLMs), trained on your private data and fully owned by you, running in your data center or on your edge workstations, at
Loop.AI
📖 Read the Full Research: The Influence of Persuasive Techniques on Large Language Models: A Scenario-Based Study By Dr. Sonali Singh & Prof. Akbar Namin from
@TexasTech
🔗
u.loop.ai/EAQYhV
#CyberSecurity #LLM #Psychology #InsideAIPodcast #LoopAI #TexasTech