Skip to content

Collections

Best Chain of Thought Episodes on AI Security

Attacking agents, defending them, and the part software cannot cover.

Charles Guillemet makes the case for hardware authorization when agents move money. Dan Lorenc unpacks vendor reports that the average time to exploit has gone negative. Kris Lovejoy discusses her forecast for autonomous attacks, Tyler Akidau examines agent governance, and Malte Ubl addresses prompt injection and tool constraints. These five conversations cover both attack scenarios and the controls guests propose to limit an agent’s reach.

  • 5 episodes
  • 5 guests
  • 4 hours of listening
  • 3 jump-to moments
  1. 1 EP 65 53 min Transcript You Can't Secure an AI Agent with Software | Charles Guillemet, Ledger Charles Guillemet, Ledger Ledger CTO Charles Guillemet argues you cannot secure an agent with software permissions and API keys once it is moving real money, and makes the case for a hardware-enforced signature between intent and execution. Jump to the moment · 9:24 · Hardware signature for authorization →
  2. 2 EP 68 47 min Transcript Time to Exploit is Negative: AI Broke the Patch Cycle | Dan Lorenc, Chainguard Dan Lorenc, Chainguard Chainguard’s Dan Lorenc on vendor reports, from Mandiant and CrowdStrike among others, that the average time to exploit is now negative, and his theory of how cheap models squeezed the long tail of late exploits. Jump to the moment · 3:42 · How time to exploit went negative →
  3. 3 EP 62 46 min Transcript The First Fully Autonomous AI Attack is Coming | Kris Lovejoy, Kyndryl Kris Lovejoy, Kyndryl In the episode published in June 2026, Kris Lovejoy predicts a fully autonomous attack within 18 months. She explains how attackers could chain low-risk vulnerabilities and what that would mean for patching priorities.
  4. 4 EP 60 51 min Transcript We Built Agents, Nobody Built HR | Tyler Akidau, Redpanda Tyler Akidau, Redpanda Tyler Akidau on the four pillars of agent governance, and why guardrails written where the agent can see them are not guardrails. Task-scoped identity is the foundation the rest sits on. Jump to the moment · 10:29 · Agent identity and accountability →
  5. 5 EP 39 54 min Transcript Vercel's Playbook for AI Agents: From Vibe Check to Production | Malte Ubl Malte Ubl, Vercel Vercel’s Malte Ubl on the security lesson people get wrong most often: prompt injection is not SQL injection, and the answer is constraining what the tool can do rather than filtering what goes in.

More collections

All collections →

New here? Start with five → Browse by topic →