Jay Chaudhry, Zscaler’s founder and CEO, laid out a sharp, technical case for why AI agents force a rethink of cybersecurity architecture; he warned agents will become the weakest link, explained how zero trust already maps to this threat, and argued that cloud-delivered exchange technology can isolate agent access while absorbing infrastructure costs—backed by recent strong financials that contrast with a battered stock price.
Most cybersecurity leaders are talking about AI agents, but Chaudhry cut through the fluff with blunt, practical language about the risks and the architecture needed to fend them off. He framed the problem as a shift from human error to machine-scale attack surfaces and demanded a security posture built for that reality. The tone was less marketing and more engineering reality check.
“A user is the weakest link. Tomorrow, agents will be the weakest link.”
He doubled down on the scale problem in plain terms: “agents run at machine speed. The numbers grow big time. They need no break, no sleep.” That single observation changes the threat calculus because the volume and persistence of agent activity can overwhelm legacy controls. The consequence is that perimeter thinking no longer matches how risk will manifest.
Zscaler’s stock has been punished hard this year and over the past 12 months, yet Chaudhry argued that the company’s technical posture matters more than short-term market swings. The firm’s architecture, he said, is engineered to treat every connection as untrusted and to broker each interaction with strict, identity-based checks. That stance is what he believes makes zero trust uniquely suited to agent-era security.
He spelled out the change in plain language: “Who is in? Who is out? Everyone is everywhere. And agents will be running from various locations,” Chaudhry said. The old model of network chokepoints and implicit trust inside the perimeter simply cannot handle distributed, autonomous agents running across clouds and endpoints. The shift demands fine-grained control over who or what touches each application.
The answer he promotes is an exchange model that validates identity and connects requests only to the specific application required. That approach isolates agents, limits lateral movement, and reduces attack surface by design. It treats agents like any other identity that must earn access for each interaction rather than inheriting broad trust.
“We have over 50 million users from over 45% of Fortune 500 companies who use it today,” Chaudhry said. “We’re simply extending that architecture to include agents.” That extension matters because customers already depend on a proven identity and exchange fabric, and extending those controls to agent traffic avoids reinventing security at scale. It reframes agents as entities to be managed rather than untouchable risks.
The core technical challenge he stressed is secure agent-to-agent and agent-to-application communication, and he was clear about how to address it. “The best approach to have secure communication among agents or agents through applications is to have an exchange technology that we pioneered,” Chaudhry added. This exchange enforces which agents can access which apps, limiting blast radius.
“This group of agents can only access this group of applications. That’s a big deal. We got lots and lots of customers waiting for this technology so they can start rolling out their agent solutions.” That customer demand, he said, is real and immediate as enterprises prototype agentic workflows that must be governed. The architecture is about enabling safe adoption, not blocking innovation.
On infrastructure costs, Chaudhry offered a pragmatic take on rising memory and processing prices and why Zscaler’s model is less exposed than hardware-centric vendors. “If we were shipping thousands of boxes like firewall companies do, it’ll be a big impact. It has some impact on us, but not in a meaningful way,” he said. The cloud-delivered setup centralizes costs across a global footprint instead of multiplying them across countless physical appliances.
Chaudhry also called frontier AI lab IPOs a net positive for his business. “Model companies going public is a good thing. It’s a tailwind for us,” Chaudhry said. “As these models are maturing, enterprises can use more and more agentic applications. More adoption of agentic technology means more need for cybersecurity. That’s where we come in.” That market dynamic ties broader AI adoption directly to increased demand for the controls Zscaler sells.
Finally, Zscaler’s recent quarter showed strong growth and profitability, with revenue up 25% year over year to $850.5 million and ARR rising 25% to $3.525 billion, while non-GAAP operating margin reached 23% and EPS climbed to $1.08. “Our differentiated Zero Trust SASE architecture, which hides applications from attackers and eliminates lateral movement, has never been more essential in securing against threats exposed by frontier models and compromised AI agents,” Chaudhry said in the earnings release. Those numbers and that technical thesis together form the argument he laid out for why zero trust should be central to securing the agent future.
