Skip to content
All writing
November 21, 20266 min read

Shadow AI and the New Job of the SSE Vendor

Someone at your company is pasting source code into a free chatbot right now. Here's who actually stops that — the SSE vendors like Zscaler and Netskope — and the distinction that separates people who've worked in this space.

  • Shadow AI
  • SSE / CASB
  • Data Security
  • AI Governance

Somewhere in your company right now, someone is pasting a customer list, a chunk of source code, or a contract into a free AI chatbot to "just clean this up real quick." That's shadow AI — employees using AI tools that security and IT never sanctioned, with data flowing to places nobody is watching.

It's the same story as shadow IT a decade ago, moving faster. And the question I get asked is: who actually stops this? The answer is a category of vendors you've probably heard of — Zscaler, Netskope, Palo Alto Networks, Cloudflare — and the name for what they do is SSE: Security Service Edge.

What SSE is, in plain terms

SSE is the bundle of controls that sit between your employees and the internet. It's usually made of a few acronyms:

  • SWG (Secure Web Gateway) — inspects and filters web traffic.
  • CASB (Cloud Access Security Broker) — sees and governs SaaS app usage.
  • ZTNA (Zero Trust Network Access) — controls access to internal apps.
  • DLP (Data Loss Prevention) — detects and blocks sensitive data from leaving.

For years this stack was about SaaS and web. Now its newest and loudest job is AI.

How SSE vendors tackle shadow AI

Because this traffic flows through them, SSE vendors can do for AI what they did for shadow SaaS:

  1. Discovery. You can't manage what you can't see. SSE surfaces every AI app and domain employees actually hit — not just ChatGPT and Claude, but the hundreds of GPT-wrapper tools and browser extensions nobody vetted.
  2. Access control and coaching. Instead of a blunt block, modern policy is nuanced: allow the sanctioned enterprise tenant, "coach" users with a warning on consumer tools, block the sketchy unknowns outright.
  3. DLP on the prompt. This is the big one. The same inspection that stopped someone emailing a spreadsheet of SSNs now inspects what's being typed into a prompt — blocking source code, secrets, PII, and regulated data before it's submitted.
  4. Tenant restrictions. Enforce that only corporate AI accounts can be used, so data stays inside contracts you control.
  5. Risk scoring. Rate AI apps by their data handling, training practices, and hosting, and drive policy from that score.

The distinction worth knowing

Here's the nuance that separates people who've actually worked in this space: SSE governs your employees' access to third-party AI. It does not secure the AI you build.

Those are two different problems with two different toolsets:

  • Shadow AI / access governance (the SSE world): protecting data as employees use outside AI tools.
  • Runtime AI security (a newer category): protecting the AI applications and agents your company ships — from prompt injection, data exfiltration, and agent misuse — at runtime.

A serious AI security program needs both. SSE is the front door that keeps your data from walking out through a chatbot. Runtime controls keep the AI products you build from being turned against you. Confusing the two is how organizations end up with a blocked ChatGPT and a wide-open internal agent.

The takeaway

Shadow AI isn't going away — the tools are too useful and the friction to use them is zero. SSE vendors like Zscaler and Netskope give you the visibility and the prompt-level DLP to govern it, which is the right first move. Just don't mistake "we govern access to outside AI" for "our AI is secure." Those are different doors — and attackers will try both.

Davis Barillas is a forward deployed engineer working across security, identity, and AI. I build small, sharp, privacy-first tools around the problems I run into in the field.