The Information Commissioner's Office (ICO) published its comprehensive guidance on generative artificial intelligence and data protection in October 2026. The guidance sets out how UK GDPR and the Data Protection Act 2018 apply across the AI lifecycle — from scraping training data to deploying customer-facing AI systems.

Lawful basis for training. The ICO confirms that legitimate interests under UK GDPR Article 6(1)(f) can provide a lawful basis for processing publicly available personal data to train AI models, but only where a genuine three-part legitimate interests assessment (LIA) has been carried out. Relying on legitimate interests without a documented LIA is unlawful. Where the training data was scraped without notice to data subjects, the ICO expects businesses to provide privacy information at first contact with affected individuals.

Purpose limitation. Personal data collected for one purpose cannot be used to train an AI system for a materially different purpose without a new lawful basis. This applies to customer data, employee data, and any other personal data collected in the course of business. Pseudonymisation reduces risk but does not render data outside UK GDPR if re-identification is reasonably possible.

Automated decision-making. Where an AI system makes decisions that significantly affect individuals (for example, credit decisions, recruitment screening, or medical triage), UK GDPR Article 22 rights apply: the individual must be told a solely automated decision has been made, must be able to request human review, and must be given a meaningful explanation. The ICO's guidance on automated decision-making sets out the compliance steps. Individuals whose data is used by AI systems may exercise their rights — including the right to a copy of their data — by making a subject access request.

The ICO has published a self-assessment toolkit for AI developers and deployers on ico.org.uk. Organisations that fail to comply with UK GDPR face fines of up to £17.5 million or 4% of global annual turnover, whichever is higher.