The S3 Brief · Issue 11 · Risk & Vendors

Your AI Arrived Through Vendors. You Don't Know Where It Is.

72% of enterprises scaled AI across operations, but only a third control it. Your vendors deployed it without you knowing.

You scaled AI across your business. Your team built pilots. You hired data scientists. Then you discovered that half your AI systems arrived through your vendors, cloud platforms, and third-party tools. You were never asked. You were never consulted. It just showed up.

This is your 2026 governance problem.

AI integration through vendors is passive and silent. Software-as-a-service platforms, cloud providers, managed service providers, and legacy software tools now include AI features by default. Your accounting system uses AI for invoice matching. Your HRIS uses AI for resume screening. Your data warehouse uses AI for query optimization. None of these required your approval.

The research confirms it. Seventy-two percent of executives report that AI has been scaled across most or all business initiatives. Only about a third say they have the right controls in place. That gap is your compliance risk.

Your compliance program doesn't see this AI. Your vendor assessment didn't ask for it. Your incident response plan doesn't address it. Your audit trail doesn't track it. From a governance standpoint, it does not exist. Until a regulator asks where it is.

Regulators are asking now. The EU AI Act requires organizations to identify and assess high-risk AI systems by December 2, 2027. Colorado passed a comprehensive AI law, though its start date has been pushed back. The SEC listed AI among its 2026 examination priorities. All of these regulations treat passive AI the same as intentional AI. The fact that a vendor deployed it without asking does not reduce your liability.

The visibility problem comes first. You cannot govern what you cannot see. Your GRC team needs a complete inventory of AI systems in use, including those embedded in vendor platforms. This means asking direct questions during vendor onboarding, reviewing contracts for AI disclosure clauses, and auditing your cloud and SaaS usage logs for AI features that are enabled by default.

When you do find it, the question is control. Does your vendor have AI governance? Does your contract include audit rights for their AI systems? Do they have documented controls for bias, security, and data handling? Most vendor contracts do not address AI specifically. They predate the technology. Your standard information security questionnaire does not ask about generative AI, model training data, or retraining schedules. That is now a compliance gap.

The second problem is scope creep. A vendor may have certified ISO 42001 compliance, but that certification applies only to the systems they explicitly declare in their management system scope. If they offer ten SaaS products and only certify three of them, the other seven are outside the audit boundary. Your purchasing team thinks you are covered. You are not.

Your controls need to live where the AI lives. If your vendor embeds AI into their product, your governance framework must extend into their product. That means contractual requirements for model documentation, testing records, incident reporting, and data retention. It means access to audit logs and risk assessments. It means the right to conduct your own assessments if regulatory scrutiny increases. Without this, passive AI becomes invisible risk.

The practical starting point is simple. Create a registry of all AI use cases across your organization. Include internal AI you built. Include AI in vendor products. Include AI in cloud services. Then classify each one: Is it high-risk under your applicable regulations? Does it handle sensitive data? Does it make decisions that affect individuals? For each high-risk system, verify that your vendor has documented controls and that your contract includes audit rights. Document all of this. Your auditors will ask for it.

Regulators are embedding AI governance into the architecture of compliance itself. Privacy is no longer disclosure. Governance is no longer annual. Compliance is now continuous visibility into systems you do not own. Vendors that have already built ISO 42001 compliance frameworks are moving faster, but certification alone is not control. The control lives in your contract, in your monitoring, and in your risk registry.

The enterprises moving fastest are treating vendor AI risk as inherent risk, not optional risk. They are asking audit questions in vendor RFPs. They are mapping vendor AI systems to their own AI inventory. They are including AI governance in their annual third-party risk assessments. This is becoming table stakes.

You have until December 2027 to demonstrate to regulators that you know where your AI is and how it is governed. Build your AI inventory now. Start with your top-ten vendors and your top-ten cloud services. Ask direct questions about embedded AI. Document what you find. This is the foundation.

If you are uncertain where to start or how to prioritize your vendor AI assessment across your organization, grab the Risk Assessment Template and adapt it for AI-enabled vendor relationships.

https://s3advisorygroup.com/resources/security-risk-assessment-template

Sources

More past issues

All issues