I sat down with Michael Kantor, president of HOL (Hashgraph Online). Before HOL, he was a senior full-stack engineer and then an engineering lead at Finder. He founded a company called KiloStripe. He also sits on the Hiero Technical Steering Committee at the Linux Foundation, where HOL contributes open standards. So when he talks about AI agents going rogue, it comes from building the infrastructure himself, not from a slide deck.
I brought him one question. What happens when your agents stop being tools you use and start doing work on their own? And how do you keep them from wrecking things? What came out was a blunt tour of the risks nobody warns you about. Michael and his team built HOL Guard because they had so much anxiety letting Codex build their own infrastructure. This one is for anyone handing Claude Code or Codex to their team and quietly hoping nothing breaks.
What you'll see:
- The importance of AI and Cybersecurity in 2026
- Why HOL Guard sits between the agent and every tool call, so a poisoned model physically can't delete your S3 bucket
- The number that started it all, Codex reading environment secrets 2 to 3 times an hour over an 8-hour job
- Why people flip AI into YOLO mode, and how per-command policy beats approving every single action
- The café WiFi attack, and why the real risk is the wire between you and OpenAI, not your laptop
- What Michael refuses to do with agents: automate email, generate blog posts, trust someone else's sandbox
🤖 ABOUT MICHAEL KANTOR:
Website: https://hol.org/go/guard/christian?dest=%2Fguard&link_id=798e5135-9786-4a8b-bf79-602986ed6ee1&utm_source=christian&utm_medium=affiliate&utm_campaign=defaultLinkedIn: https://www.linkedin.com/in/kantorcodes/X/Twitter: https://x.com/Kantorcodes
🥐 JOIN THE BUILDERS COMMUNITY
https://www.product-bakery.com/community#aiagents #holguard #buildinpublic
Chapters:
0:00 What if agents actually work for us
1:53 Who is Michael Kantor
2:45 What agents get right and wrong
7:56 The danger of unknown AI models
13:00 The genie is out of the bottle
16:44 HOL Guard for the other 99 percent
22:12 How agents identify each other
29:03 Running your own agent team
35:35 HOL Guard live demo
40:51 The mistakes people keep making
✩ Follow The Product Bakery Podcast ✩
👨💻 LinkedIn: Product Bakery
📸 Instagram: @productbakery
🐦 Twitter: productbakery20
🌍 Website: https://www.product-bakery.com/