Did Your Software Vendor Just Add AI Without Asking?

TL;DR: Yes, and it's easy to miss. Vendors are adding AI features to software your company already approved, sometimes switched on by default. One healthcare company counted and found these features were its largest source of AI requests. The fix is a named owner for each app and a re-check whenever the vendor changes something.

Nobody sends a memo that says, "We added AI to the thing you already use. Surprise." It shows up in an update. Then it's in your calendar app, your CRM, your HR system, and your email, reading data your review never covered.

Security leaders keep raising this one. It also has one of the simpler fixes.

Can software you already approved add AI on its own?

It can, and it's happening now. Approving a product means you reviewed what it did that day. When the vendor ships an update, what it does can change, including what information it reads and what it can do for your employees.

That's why an approval isn't a permanent state. It's a decision about a specific version of a product, and products keep changing.

Think of what the review checked. Did the tool see sensitive data, and could it take actions? Who was responsible for it? An AI summary or assistant can change the answer to each of those questions overnight. Your paperwork still says yes.

How common is it?

Common enough that one company's numbers are worth a look. A healthcare services company shared its count of AI requests with Josh Woodruff on a call in November 2026. Of 208 requests so far this year, 43% were AI features from vendors. AI agents were 21%.

It's one company's data, so don't treat 43% as a national average. The pattern is what's worth noticing. Vendor-added AI came up on four client calls between September 23 and November 4, 2026, from a biotech company to a benefits administrator. On the November 3 call, a security director called it the biggest shadow AI at their organization. The tool they'd bought to spot AI use hadn't caught it.

Everyone's watching for the big, dramatic AI project. The feature that appeared in an update is the one nobody put on a list.

Why does it slip past most companies?

Because nothing looks different from the outside. Same app, same login. Security tools that watch network traffic can see that people reach an AI service. They can't see that an app you approved just grew an AI feature inside it.

There's a people problem too. Reviews happen when someone asks for new software. A vendor update isn't a request, so nobody opens a ticket.

Some teams handle this with an awareness campaign, asking employees to report AI features when they spot them. It's a fair start. It also means you find out after the data has already moved.

What should a business leader ask vendors?

Ask vendors four plain things. Will you tell us before you add an AI feature? Can we turn it off until we've reviewed it? Does our data train your models? Who do we call when something changes?

Put the answers in writing, ideally in the contract or its renewal. The NIST AI Risk Management Framework Playbook asks organizations to have policies for third-party AI and plans for when a high-risk outside system fails. A vendor's AI update is a clean example of why.

Pay attention to the answer to the second one. If a vendor can't let you switch a feature off, you've learned something about how much control you really have.

Who should own the decision to turn a feature on?

One named person per app, and it shouldn't be "the IT team." Someone has to read the vendor's release notes and say yes or no. That person doesn't need to be technical. They need to know what the app is used for and what information it handles.

Josh makes the same point from the security side in his post on what happens when a vendor switches on AI in an approved app, which goes deeper on the re-review triggers and the inventory columns. For leaders, the short version is that an owner's name next to each app turns a surprise into a routine check.

What's one thing to do this month?

Pick your five most-used business apps and email each vendor the four questions above. Then write down who owns each app. That's an afternoon of work, and it gives you a record to build on.

If you're deciding what to fix first across everything AI touches, the post on which AI app problems to fix first shows how to rank by what each problem can reach.

Key takeaways

• An approval covers the version of the software you reviewed, not what the vendor adds next.

• One healthcare company found vendor AI features were 43% of its AI requests this year.

• Network tools can't tell that an approved app gained an AI feature.

• Ask vendors to notify you first, let you turn features off, tell you whether your data trains their models, and name a contact.

• Name one owner per app, and make the vendor's release notes part of their job.

FAQ

Is AI in my software the same as shadow AI?

Related, but not the same. Shadow AI is software employees bring in without approval. This version arrives inside software you did approve, which makes it harder to spot.

Do I need to stop using the software?

No. Most features are low risk and can be approved quickly. The goal is to know about the change and make a decision, so it isn't made for you.

Can I turn these features off?

Often, yes, in the admin settings. Ask the vendor where the setting is, and ask before the feature ships, not after.

Who inside the company should check this?

The person who owns the app, with help from security and privacy for anything that touches sensitive data. If nobody owns it today, that's the first thing to fix.

Software keeps changing after you say yes. Your review should keep going too, starting with whoever reads the release notes.