The threat landscape for AI systems is exploding—and runtime attacks on large language models are leading the charge. Diana Kelley, CISO at Protect AI, teams up with Chris De Vries, Senior Applied Researcher from Protect AI's Layer team, to break down the alarming surge in jailbreak techniques and denial-of-service exploits targeting production AI deployments.This conversation exposes how isolated research experiments have evolved into weaponized attack toolkits, freely shared across GitHub repositories and accessible to threat actors worldwide.
Demystifying AI Jailbreaks - Get the real story on how these attacks manipulate safety guardrails and why they represent genuine enterprise security risks, not just theoretical concerns
Beyond Jailbreaks: The Full Attack Spectrum - Uncover denial-of-service techniques, bias exploitation methods, and multimodal vulnerabilities that can cripple AI systems in unexpected ways
The GitHub Effect - See how open-source communities are turbocharging attack evolution, transforming simple prompt tricks into sophisticated, automated assault methods
Detection Reality Check - Navigate the trade-offs between catching real threats and maintaining system usability as attack patterns constantly shift
Defense Playbook for Leaders - Build effective runtime monitoring, implement layered protection strategies, and foster the cross-team collaboration essential for AI security success
CISOs, security architects, AI product managers, and development teams responsible for LLM deployments in live environments.
Diana Kelley - CISO, Protect AI
Chris de Vries - Senior Applied Researcher, Protect AI