AI Data Governance: Practical Steps for Businesses

Harper James is a UK full-service commercial law firm advising businesses on data protection, technology contracts and the laws of England and Wales. AI data governance applies those disciplines to how an organisation collects, uses, stores and shares data through artificial intelligence systems.

1. Create an AI and data inventory

Start by listing every AI system in use or under development. Include public tools used by staff, embedded features in software supplied by third parties, internally developed models and systems operated by contractors. Record the business purpose, users, data categories, supplier, hosting location, retention period and connection to other systems.

The inventory should identify personal data, confidential business information, special category data, customer records and commercially sensitive material. Each system should have a named business owner and a documented assessment of its purpose, risks and permitted uses. This process also exposes unapproved tools, duplicated systems and data flows that are not covered by existing records of processing under the UK GDPR.

2. Classify risk and set controls

Businesses should classify AI uses according to their impact on individuals, the sensitivity of the data and the consequences of an incorrect output. A system that summarises internal documents requires different controls from one used to assess job applicants, approve credit or make decisions about customers.

Controls should cover access permissions, human review, accuracy testing, audit logs, data minimisation, retention and secure deletion. Staff policies should state which information may be entered into external AI tools and which uses require approval. Where processing presents a high risk to individuals, the organisation should complete a data protection impact assessment and define how it will meet transparency, fairness, security and individual rights obligations.

3. Govern suppliers and ongoing use

AI supplier contracts should address confidentiality, data ownership, permitted processing, model training, international transfers, security measures, subcontractors, audit rights, service changes and deletion at the end of the relationship. The contract should also explain how the supplier handles prompts, outputs, customer data and system logs. Procurement teams should check whether the supplier uses submitted information to train a general model.

Governance is an operating process rather than a one-off policy. A cross-functional owner, involving legal, information security, compliance, procurement and relevant business teams, should approve higher-risk uses and review them at set intervals. Monitoring should cover inaccurate or biased outputs, security incidents, complaints, regulatory developments and changes to the system or its data. Staff training, incident procedures and periodic audits provide evidence that the organisation controls AI use in practice, not only on paper.