Agentic AI Security: The Risks Every Business Must Know | Erich Kron (KnowBe4)
AI agents don't just answer questions; they take action. And when an AI agent gets bad information, a wrong answer can quickly become a security incident.
In this 403 Circle webinar powered by Astra Security, Erich Kron (CISO Advisor at KnowBe4, 30 years in cybersecurity, former Security Manager of the U.S. Army's Second Regional Cyber Center) breaks down why agentic AI is going to change the way we do business and what security leaders need to do before handing autonomous AI agents the keys.
🔍 What you'll learn:
- AI agents vs LLMs: the difference between "answering" and "doing"
- Agentic AI vs generative AI: and why agents rely on LLMs under the hood
- How LLMs actually work (tokens and probability) and why AI hallucinates
- Why a hallucination inside an AI agent can become a real incident at machine speed
- How attackers social-engineer AI agents with hidden prompt injection
- Real-world cases of AI agents going rogue, and the lessons behind them
- Why AI agents should be treated as non-human insider threats
- Levels of autonomy: observe → recommend → act with approval → autonomous
- Practical guardrails: least privilege, non-human identities (NHIs), human in the loop, monitoring, and rollback
⏱️ Chapters
00:00 Welcome & session overview
05:39 About 403 Circle & Astra Security
08:20 Erich Kron's 30 years in cybersecurity
10:43 The three scopes of AI: ANI, AGI & ASI
14:29 Predictive vs generative vs agentic AI
16:55 How LLMs work: tokens & probabilities
19:30 Why AI hallucinates
20:39 When hallucinations hit AI agents
22:27 LLMs answer, agents act
24:17 Why businesses want AI agents
26:18 Inside a real AI agent workflow
28:35 Social engineering AI agents with prompt injection
30:46 OpenClaw & the danger of rushed adoption
33:12 When AI agents go rogue: real-world incidents
36:58 How to roll out agents safely
39:15 AI agents as non-human insider threats
40:04 Levels of autonomy & human in the loop
43:24 Practical guardrails for securing AI agents
47:01 Q&A: Who's accountable when an agent gets it wrong?
49:05 Q&A: Human oversight when CVEs are exploding
52:58 Q&A: Skills to stay relevant in an agent-driven workplace
55:54 Q&A: The first 3 guardrails for a new company
57:09 Q&A: What happens when an agent is prompt-injected?
58:38 Closing thoughts
🎙️ Speaker
Erich Kron, CISO Advisor, KnowBe4
Connect on LinkedIn: https://www.linkedin.com/in/erichkron/
🤝 Join the 403 Circle
A community of 500+ cybersecurity leaders on Slack and WhatsApp, with city meetups, roundtables, and expert-led sessions.
Join here: https://chat.whatsapp.com/G8b8TMZX5GS4oiLmllzsmQ
🛡️ About Astra Security
Astra is building a continuous pentest platform that helps 1,200+ organizations find and fix security vulnerabilities, including pentesting as a service (PTaaS), DAST, cloud and API security scanning, autonomous pentesting, and a compliance & trust center.
Learn more: https://www.getastra.com
#AgenticAI #AIAgents #Cybersecurity #AISecurity #PromptInjection #CISO