The EU AI Act's transparency obligations go live on 2 August 2026. Most UK product teams have filed them under "not our problem." That assumption is worth checking carefully.
What the rules actually require
Article 50 of the EU AI Act introduces four core labelling obligations. Providers of chatbots must disclose that users are interacting with AI. Providers of generative AI systems must embed machine-readable marks in all synthetic outputs, whether text, audio, image, or video. Deployers of deepfake systems must label that content as artificially generated. AI-generated text published to inform the public on matters of public interest must also carry a clear disclosure.
A Code of Practice from the EU AI Office sets out what "labelled" means in practice. A common visual icon, short explanatory text such as "Generated with AI," and metadata-level watermarking. The standard is multilayered: visible to users at first exposure and machine-readable for detection by downstream systems.
Penalties for non-compliance with these transparency rules sit at €15 million or 3% of global annual turnover, whichever is higher.
UK companies are not automatically exempt
The EU AI Act has deliberate extraterritorial reach. If your AI system, model, or output is placed on the EU market or used by anyone in the EU, you are in scope. Registration in the UK is irrelevant.
A UK SaaS business with European customers. A consultancy delivering AI-generated reports to EU clients. A product available in the App Store in France. All of these could fall within scope of Article 50 by August.
Many UK teams are unaware of this exposure. There is no carve-out for being headquartered in London or Edinburgh.
The case for labelling even if you never touch the EU
The UK has no equivalent mandatory labelling law. AI is regulated here through existing frameworks including UK GDPR, the Data Protection Act 2018, and the Data (Use and Access) Act 2025. The UK's pro-innovation approach relies on sector regulators rather than blanket legislation.
That does not make labelling optional in practice.
Users increasingly expect to know when content is AI-generated. Trust is a product quality. The businesses that build transparency into their workflows now will be better positioned when UK regulation eventually catches up. ISO/IEC 42001, the AI management system standard, already frames transparency as a core governance principle. Aligning with EU labelling practices is a practical way to demonstrate that maturity to clients, auditors, and regulators.
There is also a regulatory direction argument. The EU AI Act will shape global norms. US state-level AI disclosure laws are emerging. The UK is likely to follow. Building labelling into your content and output workflows now is far cheaper than retrofitting it under deadline pressure later.
AI labelling is not just a compliance checkbox. It is a signal to your users that you are building responsibly. If you are evaluating your AI governance posture ahead of August 2026, EvalSpec covers the frameworks, timelines, and practical steps your team needs.
Try EvalSpec free - evaluate your AI system across all 5 dimensions. No code required.
Start your free evaluation