One Click is All it Takes: How to Defend The Enterprise In AI-native Attacks (Live @ Black Hat '26)

Can a single click expose everything your AI assistant knows about you? At Black Hat 2026, Mark Vaitsman, Security Research Team Leader at Varonis Threat Labs, reveals how AI-native attacks transform agents into powerful tools to breach an organization's security stack. This session explores how attackers are adapting classic offensive concepts to the age of AI, from prompt injection and hidden agent-to-agent communication to novel techniques for extracting sensitive information. You'll learn about attacks like reprompt RePrompt that target personal AI assistants, and SearchLeak— an attack against enterprise copilots that demonstrates how organizational data can be exposed with minimal user interaction. Along the way, he explains the mindset shifts defenders need to make as AI becomes deeply embedded in business workflows and security programs. Chapters 00:00 Introduction: The Gap Between AI Usability and AI Security 01:04 RePrompt: One-Click Attacks Against Personal Copilot 07:58 Invisible & Repeatable AI Data Exfiltration 12:55 SearchLeak: Attacking Enterprise Copilots 16:25 SearchLeak Demo and Organizational Impact 19:23 How to Defend Against AI-Native Attacks #BlackHat2026 #AISecurity #CyberSecurity #PromptInjection #Copilot #ThreatResearch #AI #InformationSecurity #Varonis #ThreatLabs