What happens when your AI agents start making decisions faster than your security team can even see them?

In this episode, I sit down with Sunil Agrawal, Chief Information Security Officer at Glean, to unpack a shift already underway in enterprises. With predictions that 40 percent of enterprise applications will include autonomous AI agents by the end of 2026, we are moving from human-led workflows to machine-to-machine interactions at a scale most organizations are not fully prepared for.

Sunil brings a rare perspective, blending more than 25 years of cybersecurity experience with an inventor's mindset shaped by over 40 patents. What stood out to me in our conversation is how quickly the traditional security model is becoming outdated. As he explained, "autonomous agents break those assumptions because they operate across tools, varying permissions and data sources with alarming speed and autonomy." This creates what he calls the "autonomy gap," in which the CIO's drive for speed collides with the CISO's need for visibility and control.

We explore how that tension is playing out in real organizations today, and why so many are already falling behind. Nearly half of businesses still lack the AI-specific controls needed to prevent untraceable incidents, and the risks are not always what you might expect. Sunil argues that the first major rogue-agent incident is unlikely to be a malicious attack. Instead, it will come from confusion: a well-intentioned system taking the wrong action in the wrong context, with consequences that ripple across the business.

The conversation then turns practical. Sunil breaks down his AWARE framework, a structured way to introduce real-time guardrails that evaluate intent, context, and risk before an agent takes action. Rather than relying on static policies, this approach focuses on continuous runtime enforcement, where systems are constantly assessed based on behavior rather than assumptions.

What I found particularly valuable is how this moves beyond theory into something leaders can act on today. From starting with tightly scoped use cases to investing in full observability, this episode offers a clear roadmap for balancing innovation with accountability. As Sunil put it, organizations that succeed will not be the ones that move fastest, but the ones that prove trust at scale.

So how do you embrace the productivity gains of autonomous AI without opening the door to invisible risk, and are your current security models ready for a world where the "user" is no longer human?

Useful Links

Visit the Tech Talks Network Sponsor NordLayer Browser