Cross-border rules and jurisdictional friction
The UK's sector-led approach to AI, shaped in part by the UK GDPR and the Data (Use and Access) Act 2025, differs from the EU AI Act's risk-based framework. A UK-based business may still fall within the EU rules when it places an AI system or model on the EU market or when the output of its system is used in the EU. Handling information about EU-based people does not, on its own, settle whether the AI Act applies.
What to put in place
Map where your AI is developed, supplied and used, identify your role and assess which obligations apply. Review international personal data transfers separately: compliance with AI rules does not replace UK GDPR transfer safeguards.
Copyright and intellectual property in training inputs
Material that is publicly available is not necessarily free to use for AI training. Copyright, database rights, licence conditions and confidentiality restrictions can affect source documents, datasets, images and code. A supplier's claim that its model is compliant is not a substitute for understanding how your own training and fine-tuning inputs may be used.
What to put in place
Record where training material comes from, check permissions and licence restrictions and assess any applicable exception rather than assuming all ingestion is permitted. Keep rights checks distinct from the lawful basis required for personal data.
Security vulnerabilities and supplier contract gaps
Model providers, cloud hosts and software suppliers can introduce risks that standard technology contracts do not address. Weak terms may leave unclear whether prompts are retained, used for training or shared with sub-processors, how incidents are reported and what happens to your information when the service ends.
What to put in place
Review each supplier's role and agree appropriate security measures, access controls, audit rights, retention and deletion terms, incident reporting and liability arrangements. Where a supplier acts as a processor, ensure the agreement meets Article 28 requirements and check overseas access and transfers.
Shadow AI and unmonitored staff use
Employees may paste confidential business information, personal data or proprietary source code into public large language models without approval. This can take information outside your business controls, create unexpected disclosures and make it difficult to track access, retention or reuse.
What to put in place
Set clear acceptable use rules, approve suitable tools and accounts and train staff on what must not be entered. Use administrative controls where available, keep a register of approved AI uses and provide a straightforward way to report accidental disclosures.