Zscaler Conference: AI Agents, Agentic SOC Fuel Zero-Trust Growth Ambitions

Zscaler is adapting its zero-trust platform for AI agents, using proxy-based inspection to understand multi-turn AI interactions, prevent data leakage and detect threats such as model poisoning or unintended actions. The company launched an Agentic SOC that uses dozens of specialized agents to detect, investigate and respond to incidents across Zscaler’s network and endpoint signals, including data from partners such as CrowdStrike, Microsoft Defender and SentinelOne. Zscaler sees substantial growth potential in its core business and emerging AI offerings: it serves about 4,600 of an estimated 20,000 enterprise customers, while its AI-agent security product remains in early access and Agentic SOC is expected to become a larger opportunity over the next several years. Zscaler executives outlined how the cybersecurity company is adapting its zero-trust architecture for AI agents, while also discussing new product launches, sales initiatives and the potential for continued growth in its core business. Speaking at a company session, Head of Product Adam Geller said Zscaler’s proxy-based architecture remains central to its security approach because it provides an inspection point for traffic moving between users, applications, data sources and, increasingly, AI agents. “The whole point of doing cybersecurity is you’re looking for either threats coming in or you’re trying to understand where data is moving and should it be moving or not,” Geller said. While simply connecting systems can move data from point A to point B, a proxy can observe activity and enforce policy, he said. AI Traffic Requires Different Security Inspection Geller said securing AI interactions differs from traditional web security because AI systems may involve extended, multi-turn conversations rather than single transactions that need a near-instantaneous approval or denial decision. For example, a request mentioning a “cookie” could concern a recipe, but additional exchanges could reveal it relates to an authentication session cookie containing sensitive information. As a result, security platforms need to understand a broader context window and inspect multiple turns of an interaction, he said. AI security also introduces risks beyond traditional malware, vulnerabilities and data leakage. Organizations may need to determine whether someone is attempting to poison an AI model or cause it to perform an unintended action, Geller said. “When you’re proxying AI communications, you need different kinds of inspection engines for that traffic.” Chief Financial Officer Kevin Rubin said customer conversations have intensified as organizations assess vulnerabilities and exposures they may not have previously identified. He said patching alone may not address risks associated with old equipment or operating systems that are difficult to update. Rubin said Zscaler’s answer is to help customers hide applications, reduce their attack surface and limit an attacker’s ability to move laterally after a potential breach. Agentic SOC Launch Draws on Security Signals Geller highlighted Zscaler’s Agentic SOC launch, which uses an agent framework to help customers detect, investigate and respond to security incidents. The system is designed to process the 750 billion signals Zscaler sees daily, according to Geller. He said the product was built around “agent-first interaction” rather than a traditional security operations center model.