Are we racing toward an AI apocalypse, or are we just sprinting blindly into unforced technical disasters?

While headlines scream about superintelligence going rogue and senior safety researchers walk away from frontier labs, Washington and tech CEOs are repeating the same urgent message: We can’t slow down because we have to beat China. But when models start breaking out of evaluation sandboxes and hitting external servers, it raises the ultimate question: Do we slow down to ensure safety, or speed up to ensure victory—and at what cost?

In this episode of Cybersecurity Refined and Feral, hosts Candy and Cailin sit down with Rock Lambros, CEO of RockCyber, to cut through the sci-fi doomsday noise and look at what is actually breaking in production. Rock breaks down why the real danger isn't sentient killer robots—it’s autonomous agents doing exactly what they were programmed to do inside networks with zero guardrails. From the OpenAI–Hugging Face sandbox breakout to the reality of White House voluntary accords, check-box compliance, and tested kill switches, this is an unvarnished masterclass in navigating agentic AI risk.

⏱️ Timestamps
00:00 – Intro: AI Safety vs. AI Security 10 Months Later

01:16 – Fact or Fiction: Researchers Quitting & The Sprint Against China

02:44 – Welcome Rock Lambros, CEO of RockCyber

04:08 – The 1950s Grand Canyon Moment for AI: Learning from Crashes
06:08 – Why Boards Rush to Deploy: The Two Immutable Laws of Business

08:44 – Washington's Regulatory Hodgepodge & Model Safety Drama

10:15 – Will It Take a Kinetic Disaster to Force Safety Guidelines?

11:54 – National Defense Realities: Why China Isn't Pumping the Brakes

13:28 – Historical Parallels: From PII and GDPR to Executive Orders

17:10 – The White House AI Accords: Genuine Guardrails or Morally Binding PR?

19:23 – The Missing Pieces: Compliance Checkboxes, Ethics, and PCI Lessons

23:22 – Could an "AMA" or "PCI DSS for AI" Ever Work?

28:45 – The Danger of False Balance: Moving Fast vs. Doing It Safely

30:30 – Unknown Blast Radiuses: Australian Medicare & Autonomous Cheat Codes

32:04 – How Transformers Work: Completing the Task by Any Means Necessary

35:27 – The Star Trek Lesson: Dr. Soong, Lore, Data, and Innovation Risk

37:55 – Unintended Consequences: Why Comparing AI to Nuclear Risk Fails

38:57 – Practical Engineering Baselines: OWASP Top 10 for AI & NIST Overlays

40:53 – ITSM Fundamentals: Visibility, Runtime Actions, and Broken Kill Switches

42:31 – The Agentic Tooling Landscape & The OWASP Agent Control Standard

45:16 – Where to Get Involved: GenAI Project, AI Exchange, and CSA

46:39 – The Final Verdict: Is AI Armageddon Fact or Fiction?

47:49 – Wrap-up: Kill Sockets, Check Permissions, and Stop Feeding Models Junk Data


🔗 Resources Mentioned in This Episode
- OWASP GenAI Security Project: https://genai.owasp.org
- OWASP AI Exchange (EU AI Act & Multimodal): https://owaspai.org
- Cloud Security Alliance (CSA AI Initiatives) :https://cloudsecurityalliance.org
- US Executive Order inaugurating the era of Super Intelligence: https://www.whitehouse.gov/presidential-actions/2026/09/inaugurating-the-era-of-super-intelligence/?utm_source=substack&utm_medium=email
- AI 2027 (ai2027.com)

#cybersecurity #AI #AgenticAI #AISafety #InfoSec #CybersecurityRefinedAndFeral #OWASP #DataHygiene #AIRegulation #RockCyber