Frontier AI Changes the Security Timeline
Frontier AI security is moving from theory to daily planning for enterprise teams. At Black Hat 2026, Deepen Desai of Zscaler explains why advanced models are changing the speed of attacks, vulnerability discovery and response. The discussion focuses on a simple reality. Defenders now need to prepare for attackers that can reason, chain steps and move at machine speed.
Desai says organizations should rethink software development, detection and response around this shift. AI can help teams find vulnerabilities earlier in the SDLC. It can also help create remediation guidance faster. That matters because attackers can use the same class of models to find zero days and connect weaknesses into real attack paths.

Zero Trust Becomes More Important
Frontier AI security also makes zero trust architecture more urgent. Desai warns that enterprises should assume an identity, device or agent may eventually be compromised. The goal is to limit the blast radius before that compromise spreads. For Zscaler, that means hiding exposed infrastructure, reducing external attack surfaces and applying segmentation across users, devices and agents.
The conversation also covers the rise of AI agents as a new attack surface. These systems may work alongside employees and access business services. That creates new governance and monitoring needs. Traditional security basics still matter, but they must be applied to faster and more autonomous environments.

AI Defense Needs Business Context
Zscaler is applying lessons from Project Glasswing and other frontier model testing to its own security work. Desai describes AI harnesses as force multipliers, but only when they receive the right context. Models need information about architecture, threat models and business priorities to produce useful findings.
The interview also highlights exposure management, AI asset discovery, red teaming and vulnerability prioritization. Desai argues that teams cannot patch everything. They need to focus on the risks that can be chained into meaningful attacks. That requires technical insight and business context working together.

Security Teams Need to Move Faster
The larger message is practical. Security teams should use AI to improve their own workflows before attackers gain more leverage. That includes better code review, faster patch generation, stronger segmentation and improved SOC response. Desai sees this as a chance to build more secure software, but only if organizations act quickly.
For cybersecurity leaders, the interview offers a clear path forward. Reduce what attackers can see. Contain what they can reach. Use AI to find and fix the highest-risk issues. Then build security programs that can respond at the same speed as the threats they face.