Third-party and embedded AI

The Uninvited Guest

Ask a mid-sized business whether they use AI and a good number will say 'no, not yet, we're looking at it.' They aren't lying; they're just wrong. The AI has already arrived inside products they already bought, and nobody sent a memo. Think about the ingredients in a supermarket ready meal. When you buy a frozen lasagna, you're choosing the meal, but you're also getting whatever ingredients the manufacturer decided to put in it—like palm oil or thickeners. You didn't specifically set out to buy those individual components, but they're in your kitchen now anyway. Software works the same way.

Procured vs. Inherited AI

We need to draw a sharp line between two ways AI enters a business. First, there's Procured AI. This is a deliberate choice. The firm identifies a need, evaluates a tool, and signs a contract. It's visible and inventoried. Then there's Inherited or Embedded AI. This arrives uninvited inside software the business already runs. Because it arrives as a 'new feature' in a familiar tool, it often bypasses the usual procurement and risk assessment hurdles.

The Update Trap

Embedded AI refers to capabilities tucked inside a product bought for another purpose. It often arrives via a routine update mid-contract. An employee might see a new 'Summarize' button in their email client and start using it immediately. Nobody in IT or Risk was ever notified that the firm's data is now being processed by a large language model. The exposure arrived automatically, without a single signature.

Recap: AI Entry Points

Let's pause and recap. We've seen that most AI exposure isn't a choice—it's a side effect of the software you already use. Whether it's procured or inherited, knowing the difference is the first step in identifying risk.

Why the Inventory Question Fails

If you ask a mid-sized business whether they use AI, many will say, 'No, not yet.' They aren't lying; they're just wrong. They're looking for AI as a standalone product, but they're missing the AI already running in their spreadsheets, their meeting recorders, and their HR portals. This creates a massive gap in the AI inventory—the record of where AI is actually in use. The person answering your question genuinely doesn't know it's there because it arrived through a channel they don't think twice about.

Vendor Dependency & Concentration

This creates vendor dependency. The customer inherits the vendor's failures—like fabrication or bias—but has zero visibility into how the model works. Even more concerning is that many different software vendors actually sit on the same few underlying 'foundation' models. This means a single failure at a tech giant could ripple through hundreds of seemingly unrelated business tools simultaneously. We'll explore this 'concentration risk' more in lesson ten.

Common Pitfalls

To spot real exposure, you have to avoid three common pitfalls. First, the 'No AI' Trap—assuming no risk because no standalone product exists. Second, Label Reliance. Many vendors hide AI under names like 'Smart Assist' or 'Magic Tools.' Finally, Static Inventories. Because vendors add features mid-contract, an assessment from last quarter may already be wrong.

Recap: The Inventory Gap

Before we move to practice, remember: the inventory gap is real. Don't just look for labels; look for how work is getting done.

Asking the Right Questions

A client tells you they don't use AI. You need to uncover the real risk. Which follow-up question is better? Type your choice (A or B) and explain why. Let's see how you did.

The Explicit Stop

We've covered why organizations have AI exposure they never bought. You've seen how updates and embedded features create a moving target for risk assessment. Building a real inventory, and putting the right questions to a vendor, is the next course. Here it is enough to stop believing the answer 'we don't use AI', including when your own firm gives it.