Zscaler AI Governance: Why AI Security on Zscaler Is a Stack of Add-On SKUs

Zscaler is the category leader, and it earns that on scale. But when a team asks the simple question, can we allow our corporate ChatGPT and block the personal one while inspecting what goes into the prompt, the answer on Zscaler is rarely a clean yes. It is a yes, with the Data Protection add-on, on the Transformation or Unlimited edition, plus a couple of other modules. This is a Zscaler displacement story for buyers who want AI governance without assembling a license stack. For the full picture, start with our complete guide to replacing Zscaler.

What does AI governance actually cost on Zscaler?

The honest answer is that it depends on how many add-ons you already own, and most teams do not own all of them. Zscaler ships strong AI discovery, so seeing AI usage is the easy part. The expensive part is acting on it. Prompt level DLP requires the Data Protection add-on. AI Guard and the AI Scanning Platform are separately licensed. User coaching, the warn and educate workflow, leans on Workflow Automation. Each of those is a line item layered onto the base proxy, and the base proxy itself comes in stacked editions, Business, Transformation, and Unlimited, where the AI relevant capabilities tend to live in the higher tiers.

That structure is the point. Zscaler is documented as delivering AI controls through add-ons rather than as a native function of the gateway. So the sticker you compare against a competitor is not the sticker you pay once you add the pieces that make AI governance real. Our Zscaler pricing comparison breaks down how those editions and add-ons stack in practice.

The renewal is where the math gets uncomfortable

Pricing pain on Zscaler concentrates at renewal, and it is documented. NPI Financial reported in August 2025 that some Zscaler SKUs were running 35 percent or more above prior pricing, without a public announcement. Stack that on top of the per edition, per add-on model and the AI governance line items, and the number that looked competitive at signing looks very different three years in. Customers report that the editions and modules are hard to map to what they actually use, which is exactly the condition under which renewal increases are hard to challenge.

None of this is a knock on whether Zscaler works. It is a knock on the buying model. You are paying for AI governance as a series of upgrades to a proxy, and the proxy is the thing adding cost and latency in the first place.

Why the proxy architecture taxes AI workflows

Every request on Zscaler forwards to a service edge node, because the proxy sits in the data path. That is how a cloud proxy works, and it is also why latency compounds as you stack modules. Gartner has cited a throughput drop in the 10 to 20 percent range, and customers report 2x to 3x latency as inspection layers add up. AI workflows are chatty, with rapid back and forth between the user and the model, so the latency tax is felt more here than on a static web page.

There is also a reliability dimension that matters for anything you depend on daily. Zscaler has a documented pattern of outages tied to its own maintenance. On October 25, 2022, a subset of ZIA proxies saw total packet loss after internal maintenance hit its own node addresses. On January 19, 2025, a multi service outage occurred during scheduled maintenance, and Dark Reading framed it as a redundancy problem. When the control plane has a bad day, the dashboards and logs you would use to manage AI policy can go with it. Our breakdown of why teams are replacing Zscaler covers the operational pattern in more detail.

Does Zscaler break the apps your AI users rely on?

Sometimes, and it is documented behavior, not a fluke. Zscaler cannot inspect certain cert pinned applications, including Microsoft 365, WebEx, and Dropbox, which pushes teams to build bypass lists. Those bypass lists become blind spots, and AI traffic that rides through a bypassed app or a pinned client slips past inspection. Zscaler own documentation also tells admins to watch for high CPU and memory growth, and the client connector can hang for up to 60 seconds on a wired to Wi Fi transition. For a workforce that lives in browsers and desktop AI clients, that friction is a daily tax.

dope.security versus Zscaler for AI governance

The contrast is architectural. dope.security runs a lightweight agent on the device, inspects SSL locally, and flies traffic direct to its destination with no backhaul. AI governance is native, not an add-on: shadow AI discovery, Cloud Application Control for tenant level allow and block on the same domain, and Dopamine DLP for prompt and upload inspection with zero retention classification. One console, one agent, under 100 MB of RAM, and roughly 4x the performance of legacy proxy SWGs.

What switching actually looks like

The fear with any displacement is deployment lift, and this is where the agent model wins. A Fortune 100 company rolled dope.security out to more than 18,000 devices in record time, and Outreach Health secured 99 percent of its devices within a week while cutting web access tickets by 70 percent in 90 days. There is no data center setup, no PAC file surgery, and no months long implementation, because the security runs on the endpoint and policy pushes in seconds. You move from a proxy you keep upgrading to a platform where AI governance is already included.

The bottom line

Zscaler scores well as a proxy, and its AI discovery is genuinely strong. The problem is the model: real AI governance, the prompt inspection and the corporate versus personal tenant control, lives in add-ons and higher editions stacked on the gateway, and the cost shows up sharpest at renewal. If you would rather buy AI governance as a built in capability than as a stack of SKUs on a proxy, dope.security delivers discovery, tenant control, and prompt DLP natively on the device. See it for yourself with a 20 minute demo, and read the complete guide to replacing Zscaler for the full migration path.