Third-Party Risk and AI Vendors — What CISOs Must Include in Vendor Due Diligence

The explosion of AI tool adoption across enterprises has created a third-party risk management challenge that most organisations are not yet equipped to handle. Security teams that have mature processes for assessing traditional software vendors, cloud providers, and managed service providers are finding those processes inadequate when applied to AI service providers — companies whose products ingest, process, and potentially retain sensitive organisational data in ways that are fundamentally different from conventional software.

Why AI Vendors Are a Different Risk Category

Conventional third-party risk assessment asks: what data does this vendor access, how do they protect it, and what controls do they have in place? These questions remain relevant for AI vendors, but they are insufficient. AI vendors introduce additional risk dimensions: Is customer data submitted to the AI system used to train shared models? Can other customers of the same AI service extract information derived from your data through the model’s outputs? How are AI model updates governed, and what change management controls exist? What happens to submitted data if the vendor is acquired, fails, or is compelled by a foreign government? These questions do not appear in standard vendor questionnaires — they need to be added.

PCI DSS and AI Vendor Risk

PCI DSS v4.0 Requirement 12.8 imposes clear obligations on organisations regarding their service providers. Any vendor that has access to or could affect the security of cardholder data — including AI tools used in customer service, fraud detection, or payment processing workflows — must be subject to formal due diligence, contractual security requirements, and ongoing compliance monitoring. If an AI tool is capable of processing payment card data, it falls within PCI DSS scope, and the organisation is responsible for ensuring that the vendor maintains adequate security controls. Gaps in service provider management are commonly identified during security assessments and should be addressed proactively.

ISO 27001 and Supply Chain Security

ISO 27001:2022 strengthened supply chain security requirements, with controls addressing information security in supplier relationships. For AI vendors, this means due diligence must extend beyond the vendor itself to include the security posture of its underlying technology providers. This includes cloud infrastructure providers, model repositories, and third-party APIs that support the AI service. Organisations should ensure supplier agreements include clear clauses covering data handling, incident notification, and verification rights.

SOC 2 and AI Vendor Assessment

SOC 2 reports are a useful indicator of a vendor’s security posture, but they are not sufficient on their own when evaluating AI services. It is important to review the scope of the report carefully. Does the report include AI-specific components, or only the underlying infrastructure? Are there exceptions related to data protection or privacy? A clean SOC 2 report with a limited scope does not provide full assurance for AI-integrated systems.

Building an AI-Aware Vendor Risk Process

CISOs should update vendor risk management frameworks to include AI-specific evaluation criteria. This includes maintaining an inventory of AI tools, classifying them based on data sensitivity, and implementing tailored due diligence processes. Additional measures should include AI-focused questionnaires, policies governing data usage, and continuous monitoring mechanisms. This is not a future requirement — it is an immediate need as AI adoption accelerates.

Contractual Protections That Matter

Strong contractual safeguards are essential when working with AI vendors. Agreements should clearly define how data is used, stored, and protected. Key provisions include restrictions on using client data for model training without consent, requirements for data deletion, rights to verify compliance, and clear incident notification timelines. Vendors unwilling to agree to such terms should be considered higher risk.

Talk to Our Experts | Aegisra Assurance LLP can review and strengthen your third-party risk management programme for AI and technology vendors. Contact us to close this gap in your compliance framework. | www.aegisraassurance.com