An Information Security Vendor Assessment Framework for AI Procurement

Sammanfattning

The accelerated adoption of AI-powered products and services is transforming how organisations should procure and integrate external software. Unlike conventional software, AI products may process data through opaque inference pipelines, rely on third-party foundation models, and change behaviour through silent model updates. While standards like ISO/IEC 27001 require organisations to assess supplier risks, its controls are technology-neutral and do not guide the assessor toward AI-specific risk areas. Existing AI governance standards and industry guides describe relevant risks but do not provide structured assessment criteria that information security professionals can apply within their existing supplier evaluation process. We present a vendor assessment framework that enables information security professionals to assess AI products within an ISO 27001-aligned supplier assessment process. The framework classifies AI products along two axes, operational autonomy and data sensitivity, to determine whether a baseline or deeper diligence assessment is required. It defines 18 baseline and 13 deeper diligence assessment questions, each mapped to the CIA triad and supplemented with evaluation guidance designed for assessors without AI-specific expertise. The framework was developed using the Design Science Research method and evaluated through illustrative scenarios applying it to two real AI products with different risk profiles. Our evaluation shows that the classification mechanism produces meaningfully different assessment processes for products at different autonomy levels, and that the assessment questions surface crucial AI-specific risks.