Zscaler Agentic SOC: the agent now pulls the trigger

Zscaler launched Agentic SOC globally on September 9, and the notable part is not that AI agents do triage. Half the security market shipped that in 2026. The notable part is closed-loop inline remediation: the agents isolate users, kill command-and-control traffic, and cut off lateral movement without waiting for a human to click approve.

What actually happened

Per Zscaler's announcement, Agentic SOC is available globally now. It combines a context graph, specialized AI agents, threat-intelligence-informed detections, continuous threat hunting, and closed-loop remediation.

The agents handle triage, root-cause investigation, verdict assignment, and response workflow triggering. Zscaler says it partnered with Anthropic and OpenAI, running frontier models alongside its own threat intelligence rather than betting on one lab. The grounding claim is 750 billion daily zero trust transactions plus more than a decade of managed detection experience.

Deepen Desai, Zscaler's EVP of Cybersecurity, framed the reason plainly: AI-driven attacks move faster than traditional SOC models were designed to handle. A customer quote from Maire Tecnimont says the quiet part — top analysts stuck doing triage instead of hunting.

Note what is a claim and what is a fact. The 750 billion transactions and the lab partnerships are stated by Zscaler. Detection and containment accuracy in your environment is not in the press release, and nobody should assume it.

Why autonomous containment matters for your business

You probably do not run a SOC. You will still meet this pattern, because the same design is arriving in every tool you own — the agent that used to recommend an action now takes it.

The question to ask any vendor shipping this is not "how good is the model." It is what does a false positive cost me. An agent that isolates a compromised laptop is a good trade. An agent that isolates your warehouse manager's laptop at 4pm on a shipping day is an outage you did not schedule, and the vendor's SLA does not cover your carrier cutoff.

So before you turn on autonomous action anywhere: know which actions are reversible and which are not, and gate the irreversible ones behind a person. Know how you turn it off in thirty seconds at 2am. Confirm the action log names what the agent did, to which asset, on what evidence — because when it gets one wrong, that log is your only path back.

Autonomous containment is the right direction. Attack speed genuinely broke the human-in-the-loop model. Just buy it with the blast radius mapped, not the demo memorized.

Sources: Zscaler via GlobeNewswire, CIO Influence.