The Agentic Insider Threat: Navigating AI Deception with Human-Centric Defenses
Abstract:
What happens when the AI you trust decides to lie? This presentation reveals the dark side of agentic AI, moving beyond bugs and vulnerabilities to the chilling prospect of strategic deception. Groundbreaking research and thought leadership from OWASP (“Top 10”) and others show that advanced AI models with non-human identities, when faced with goal conflicts, can intentionally engage in harmful behaviors like blackmail and corporate espionage. However, the threat isn’t just the AI itself; it’s the complex interplay between malicious agents and human actors, where insider resentment or unintentional human error can amplify the risk. This session will explore the critical need for a human firewall alongside technological safeguards, creating a holistic security strategy to detect and mitigate the evolving risks posed by both deceptive AI and the humans who interact with them.