AI Supply Chain Risk: What Your Vendors Are Not Telling You

When organizations think about AI risk, they tend to focus on the AI systems they are building or directly deploying. But for most small and mid-sized companies, the more immediate exposure is not the AI they are building — it is the AI embedded in the tools they are already using.

Your CRM may be using AI to score leads and predict churn. Your HR platform may be using AI to screen resumes. Your customer support software may be routing tickets and drafting responses using a large language model. Your cloud infrastructure provider may be using AI to detect anomalies and manage capacity.

Each of these systems represents an AI supply chain dependency — and most organizations have no systematic way to assess the risk those dependencies introduce.

Why Standard Vendor Risk Programs Miss AI Exposure

Traditional vendor risk management was designed for a world where the primary concerns were data handling, access controls, and business continuity. A standard security questionnaire asks whether a vendor encrypts data in transit, whether they have SOC 2 certification, and whether they have a disaster recovery plan.

These are still important questions. But they do not capture the risks that are specific to AI systems.

AI introduces a different category of exposure:

Model training data risk. If a vendor's AI model was trained on data that included your organization's information — or data from your industry — you may have exposure you are not aware of. Some AI vendors use customer data to improve their models by default. The opt-out is often buried in the terms of service.

Output reliability and drift. AI models degrade over time as the world changes and the distribution of inputs shifts. A vendor whose AI was highly accurate at deployment may be producing unreliable outputs today. Standard vendor assessments do not include model performance monitoring.

Prompt injection and adversarial manipulation. AI systems that process user-generated input — customer messages, documents, emails — can be manipulated through carefully crafted inputs designed to make the model behave in unintended ways. This is a class of vulnerability that most vendor security questionnaires do not address.

Regulatory compliance of AI outputs. If a vendor's AI is making or influencing decisions about your customers — credit decisions, insurance pricing, hiring recommendations — you may have regulatory exposure under fair lending laws, anti-discrimination regulations, or emerging AI-specific legislation, regardless of whether you built the model yourself.

Sub-processor AI dependencies. Your vendor may be using a third-party AI provider to power their product. That sub-processor relationship creates a chain of dependencies that your standard vendor review may not surface at all.

Building an AI Vendor Risk Inventory

The first step in managing AI supply chain risk is knowing what you have. This sounds obvious, but most organizations significantly underestimate the number of AI-enabled tools in their environment.

Start with a structured inventory exercise:

Identify all SaaS tools in your environment. Work with procurement, IT, and department heads to build a complete list of software subscriptions. Shadow IT — tools adopted by individual teams without formal procurement — is a significant source of untracked AI exposure.

Classify each tool by AI usage. For each tool, determine whether it uses AI or machine learning in any capacity. Vendor documentation, privacy policies, and terms of service are the starting point. Direct inquiry to the vendor is often necessary.

Assess data exposure. For each AI-enabled tool, determine what data the tool processes and whether that data is used to train or improve the vendor's models. Pay particular attention to tools that process personal data, health information, financial records, or proprietary business information.

Map sub-processor relationships. Ask each AI-enabled vendor to identify the AI providers or models they rely on. A vendor using OpenAI, Anthropic, or Google as a sub-processor has a different risk profile than one running proprietary models on dedicated infrastructure.

The Questions Your Vendor Questionnaire Is Missing

Once you have your inventory, the assessment process needs to go beyond standard security questionnaires. Here are the questions that matter for AI-specific risk:

• Does your product use AI or machine learning? If so, describe the use cases.
• Is customer data used to train or fine-tune your AI models? What is the opt-out process?
• What third-party AI providers or models does your product rely on?
• How do you monitor AI model performance over time? What is your process when model accuracy degrades?
• How do you test your AI systems for bias, fairness, and discriminatory outputs?
• What controls do you have in place to prevent prompt injection or adversarial manipulation?
• What is your process for notifying customers of material changes to your AI systems?
• Do you have documentation of your AI governance program? Are you pursuing ISO/IEC 42001 certification?

Most vendors will not have complete answers to all of these questions. That is useful information. The quality and completeness of a vendor's response tells you a great deal about the maturity of their AI governance program.

Contractual Protections That Actually Matter

Standard vendor contracts were not written with AI in mind. When negotiating or renewing agreements with AI-enabled vendors, there are specific provisions worth pushing for:

Data use restrictions. Explicit contractual language prohibiting the vendor from using your data to train or improve their AI models without your consent.

Model change notification. A requirement that the vendor notify you of material changes to the AI models powering their product, with a reasonable notice period.

Audit rights. The right to request documentation of the vendor's AI governance practices, including bias testing results and model performance metrics.

Regulatory compliance warranties. Representations that the vendor's AI outputs comply with applicable laws and regulations, including anti-discrimination requirements.

Incident notification. Explicit coverage of AI-specific incidents — model failures, adversarial attacks, data poisoning — in the vendor's breach notification obligations.

Connecting AI Supply Chain Risk to Your Broader Governance Program

AI supply chain risk does not exist in isolation. It connects directly to your data governance program, your vendor risk management program, and your regulatory compliance obligations.

Organizations that are pursuing ISO/IEC 42001 certification or building programs aligned to the NIST AI Risk Management Framework will find that third-party AI risk is explicitly addressed in both frameworks. ISO/IEC 42001 includes specific requirements for managing AI-related risks in the supply chain. The NIST AI RMF's GOVERN and MAP functions both address third-party AI dependencies. Our resource article on ISO/IEC 42001 and the EU AI Act covers how these frameworks approach supply chain risk in more detail.

If you are building an AI governance program from scratch, third-party AI risk is a good place to start — because it surfaces exposure you already have, in tools you are already using, without requiring you to build anything new. For a deeper look at the specific contract gaps that create the most exposure, see our resource article on AI supply chain risk and the contract gaps your vendor agreements miss.

Once you have mapped your AI vendor exposure, the next challenge is communicating that risk to leadership in terms they can act on. Our post on how boards should think about AI security covers the governance questions directors should be asking — and our post on quantifying AI risk with FAIR covers how to translate vendor risk into financial terms for the CFO and executive team.

GRC2 helps organizations build AI vendor risk programs that go beyond standard questionnaires to surface the exposures that actually matter. For more on our AI governance approach — including frameworks, risk quantification, and board reporting — visit our AI Governance hub. If you want to understand your AI supply chain risk, schedule a call with our team.