In this CISO Tradecraft episode, host G Mark Hardy interviews recovering CISO Rock Lambros (Zenity) about securing Agentic AI and the emerging risks beyond LLM hallucinations. Lambros recounts his path from Oracle developer to CISO and AI standards work, then explains how agentic AI increases risk by connecting models to tools and actions. They discuss agentic AI supply chain attacks, including backdoored LiteLLM packages on PyPI and a compromised Amazon Q update, and the resulting shift from “patch fast” to more cautious dependency controls. The conversation highlights the OWASP Top 10 for Agentic Applications 2026, covering threats like goal hijack, tool misuse, identity/privilege abuse, memory/context injection, insecure inter-agent communication, cascading failures, human trust exploitation, and rogue agents, concluding with practical steps: inventory, kill switches, least agency, intent gates, and observability.

OWASP Top 10 for Agentic Applications -

https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/