AI vendor security due diligence
Questions and evidence for assessing AI suppliers across data, models, identity, contracts and exit risk.
Follow the information
Ask what data enters the service, where it is processed, whether it trains provider models, how long it is retained and which subprocessors receive it. Match evidence depth to data sensitivity and business impact.
Review tenant isolation, encryption, identity, administrative access, logging and deletion.
Assess the AI-specific failure modes
Consider prompt injection, insecure output use, excessive agency, model or retrieval poisoning, unpredictable behaviour and dependency on external models. For agents, document every tool and action boundary.
Want this assessed against your environment?
Send us the specifics and a senior advisor will respond within one business day.
Make assurance contractual
Include incident notification, material model or subprocessor changes, audit evidence, service continuity, data return and deletion, transition support and liability. Reassess high-impact suppliers on a defined cadence.
Sources and further reading
- ISO/IEC 42001 AI management systems
- NIST AI Risk Management Framework
- ISO/IEC 27001 information security management systems
This article provides general information and decision support. It is not legal advice, audit assurance, certification advice or a guarantee of outcome.
Related reading
A risk-based cybersecurity roadmap for SMEs
Build a sequenced cyber programme around business exposure rather than an unprioritised control list.
Read articleISO 27001 vs Essential Eight for Australian SMEs
How the management-system and technical-control approaches differ, overlap and can work together.
Read articleReporting cyber risk to a board without technical noise
A board reporting structure centred on exposure, decisions, evidence and accountable action.
Read article