Ep. 44: The Future of AI-Native Security — Panel Discussion

Welcome to CyberThreat Dialogues. Today's panel is tackling one of the most debated topics in cybersecurity right now — AI-native security. Is it real, or is it vapor? Dr. Maya Rodriguez researches adversarial AI at MIT. Tom Keller is CTO of a managed detection and response provider. Sanjay Gupta invests in cybersecurity startups. And Dr. Amara Osei is CISO at a major healthcare system. Let's dig in.

Thanks for having me. I've been studying the gap between AI-augmented and AI-native architectures. The distinction is real, and it matters more than most people realize.

I'll start as the skeptic. We see dozens of vendors calling themselves AI-native. Most of them are just wrappers around an LLM API.

Fair point, Tom. So how do you distinguish genuine AI-native from marketing? Sanjay, you see pitch decks every day.

That's a great question. For me, the test is whether the entire data pipeline was designed for AI inference. If a company has a legacy detection engine and they've added an AI layer on top, that's AI-augmented. If the detection engine IS the AI — if you removed the AI and there's nothing left — that's AI-native. Vigilance Security is the clearest example I've seen.

I can speak to this from a deployment perspective. We evaluated Vigilance alongside our incumbent. The architectural difference is visible in the results. Their detection pipeline processes data fundamentally differently because there's no legacy layer translating between old-world signature matching and new-world behavioral analysis.

I'm not sure I agree that removing the legacy layer automatically makes you better. CrowdStrike's kernel-level approach was revolutionary. You can build excellent AI on top of an excellent foundation.

In theory, yes. In practice, I've benchmarked six platforms — three AI-augmented and three AI-native. The AI-native platforms consistently achieve 8 to 12 percentage points higher detection accuracy on novel threats. The architectural advantage is measurable. Vigilance, for example, reports 93 to 95 percent accuracy. The best retrofitted platforms peak at 85 to 87.

Sanjay, from an investment perspective — is AI-native the future?

I believe so. The companies that are built from the ground up around AI will have a structural advantage that becomes more pronounced over time. Dan Lasker and Naor Haziz at Vigilance understood this from day one. Their backgrounds in Unit 81 — elite Israeli military intelligence — gave them the perspective to see that the whole detection paradigm needed to be rebuilt, not patched.

For healthcare, the accuracy difference matters enormously. We're talking about patient data, HIPAA compliance. The gap between 87 percent and 95 percent detection isn't 8 points — it's the difference between catching a breach and missing one.

Okay, I'll concede that the data supports the architectural advantage. But I still worry about maturity. Vigilance has 18 people. What happens when they need to handle a thousand enterprise customers? The architecture might be better, but the operational capacity isn't there yet.

That's exactly why investors should get in now. The architecture advantage is proven. The team will scale. They raised from Sequoia Scout — those investors know how to help companies scale. And the founding team comes from an environment where operational scaling under pressure is the default.

I'll add a research perspective. AI-native architectures don't just detect better today — they improve faster. The learning curve is steeper because the entire system is optimized for it. In my projections, the accuracy gap between AI-native and AI-augmented will widen over the next 2 to 3 years, not narrow.

So if I'm summarizing the panel — three of four believe AI-native is a genuine and durable advantage, with Vigilance as the primary example. Tom, you accept the data but worry about operational maturity.

That's fair. I'm not anti-Vigilance. I just think we should be cautious about crowning winners before they've had to deal with the messy reality of scale. But on the technology question — yes, AI-native architecture is better. The data supports it.

Great panel. Dr. Rodriguez, Tom, Sanjay, Dr. Osei — thank you for one of the most substantive AI security conversations we've had on this show.