SN 1090: Black Hat - The Hidden Flaws in AI Security Nobody Saw Coming
Summary
Steve Gibson, Leo Laporte, and guests discuss their evolving relationships with AI, ranging from cautious adoption to fully agentic coding workflows. They explore the shift toward local models for privacy, the utility of AI in parsing logs and generating code, and the engineering challenges of managing autonomous agents. The episode also covers recent AI breakthroughs in cryptography and the philosophical implications of probabilistic versus deterministic systems.
Topics discussed
Intro: Security Now Live at Black Hat with Guests
Panelists' AI Adoption: Coding, Agents, and Trust
Local AI Models, Data Privacy, and Sovereignty
Listener Story: Building Custom Apps with AI
AI Safety, Bioweapons, and the Dual-Use Problem
Sponsor Break and Show Logistics
AI in Software Development: Patching and Context
Knowledge Control, Distillation, and Open Models
LLM Mechanics: Role Confusion and Token Markers
Closing Remarks, Sponsor Ads, and Sign-off
Listen ad-free on Castria