Your vendors' AI is now part of your risk surface
The AI in your supply chain rarely arrives with a purchase order. It arrives as a feature switched on inside software you already use — and its risk lands on you.
A few years ago, adopting AI was a decision. Someone chose a model, signed a contract, and stood up a project. Today it is more often a default. A tool you already licence ships a new assistant, a summariser, or an automated decision step, and it is on by the time anyone in risk hears about it.
That changes the shape of the problem. The AI you most need to understand is frequently AI you did not procure — it belongs to a vendor, and it is running against your data.
Why third-party AI is different
Traditional vendor risk asks about a supplier's security and handling of your data. AI adds questions those reviews were not built to answer:
- What does the model do with the data it sees? Is your content used to train or improve a model, and can you turn that off?
- How are its outputs governed? Is there human oversight on decisions that affect people, and a way to contest them?
- Whose model is it, really? Many products are a thin layer over a foundation model from another provider. The transparency you need may sit one supplier further back than your contract reaches.
- What happens when it is wrong? Accuracy, bias and failure handling are control questions now, not research curiosities.
None of these are answered by a SOC 2 report alone. They sit in a category of their own.
The frameworks are catching up
Two reference points are worth knowing, described plainly and without overstating them.
ISO/IEC 42001 is an international standard for an AI management system. Like ISO/IEC 27001 for information security, it sets out how an organisation should govern AI across its lifecycle — and, being certifiable, it gives you something a third party can attest to rather than merely assert.
The NIST AI Risk Management Framework is a voluntary framework, organised around the functions of governing, mapping, measuring and managing AI risk. It is guidance rather than a certification, useful as a common vocabulary for asking a vendor how they manage risk.
Neither is a checkbox that makes a vendor safe. They are structures for asking better questions — and for recognising a serious answer when you get one. For a definitive reading of how either applies to your obligations, involve a qualified professional; the point here is simply that vendor AI is now assessable against recognised ground.
Assessing it without a six-week project
The practical barrier is effort. You cannot run a full engagement on every supplier that switched on an AI feature. This is exactly where an outside-in view earns its place: a fast, consistent read of a vendor's public AI posture — model and provider transparency, governance, independent assurance evidence and more — graded the same way every time, with each finding linked to its source.
It will not replace a proper assessment where the stakes justify one. It will tell you, quickly, which vendors clear the bar on the public record and which ones owe you a conversation before their AI touches anything that matters.
Know what you're relying on. Run a free AI Trust Scan on a vendor in your supply chain, or commission independent assurance when the decision warrants it.