In most enterprises, AI arrived through the side door — inside the products companies already buy. A support tool added a model. An analytics vendor started training on customer data. A SaaS platform quietly shipped an agentic feature in a routine release. No new contract, no new review, just new risk folded into a relationship that was already approved.
Our survey of 309 IT and security leaders and practitioners suggests most programs can't see it. AI-in-vendors is the fastest-growing gap in third-party risk today, and it's widening precisely because the assessment model wasn't built to catch it.
You Can't Assess What You Can't Find
The gap starts with visibility. Only 37% of respondents are highly confident they can even identify where AI is being used inside a vendor's products and services. Nearly two-thirds are something less than highly confident — which means, for most programs, that AI is entering the environment through third parties faster than the team can reliably spot it.
This is a tougher problem than the first wave of shadow IT. AI capabilities show up as features inside tools you've already sanctioned, which makes them far harder to catch than a new login to block. The vendor was vetted; the model wasn't there yet, or wasn't disclosed, or arrived in an update nobody re-reviewed. Visibility here is a moving target that never fully resolves.
Developing A Consistent Evaluation Standard
Visibility is only the first step. Once you know a vendor uses AI, you have to evaluate what that means for your risk — and here the picture is thinner still. Only 42% of organizations have a standard process for assessing third-party AI risk. A majority, 53%, handle it case by case, ad hoc, with whatever questions the reviewer thinks to ask that day. A small share does nothing at all. Ad hoc or improvised evaluation is how blind spots form.
Without a consistent set of criteria — how the vendor governs its models, what it does with your data, how it handles training and retention, what happens when the model behaves unexpectedly — every review is improvised, and the results can't be compared, trended, or defended. That improvisation strips away the criteria-based rigor teams apply to the rest of their vendor risk, and it does so for the category changing fastest.
Teams know it, too. AI-specific governance is one of the weakest capabilities they rate in themselves: 38% score their ability to collect AI-governance evidence below "good." Teams see this gap clearly, but they haven't had the tools to close it.
The Demand Signal Is Already Loud
The clearest sign of how much this matters is where teams are putting their improvement dollars. Increasing their ability to assess AI-specific governance ranked as a top-two priority for the year ahead, cited by 47% — essentially tied with automating data collection at the top of the list. Out of everything a third-party risk program could invest in next, closing the AI gap is near the front of the line.
That's a striking place for a risk category to sit when a standard process for it exists in fewer than half of programs. It tells you the market has recognized the exposure well ahead of the tooling that addresses it.
Uniting AI With Everything Else
The fix is to fold AI risk into the same continuous, criteria-based assessment that governs the rest of the vendor relationship — detecting where AI is in use, evaluating it against consistent standards, and keeping that evaluation current as vendors change what their models do. A separate, bolted-on AI questionnaire only ages the moment it ships. AI stops being a blind spot the moment it's assessed the way everything else is: systematically, on defined criteria, and often enough to keep up.
The full research report covers the AI-in-vendors gap alongside the broader capacity and coverage findings shaping third-party risk in 2026.
Read The State of Third-Party Risk Management 2026, or book a demo to see how Drata assesses AI risk inside third-party products with a standardized, criteria-based process.