Starting August 2, 2026, the EU AI Act's Article 50 makes AI labelling a legal requirement for any company serving EU citizens, regardless of where the company is based. Labelling applies narrowly to deepfakes, chatbots/AI agents, fully AI-written text on matters of public interest with no human review, and emotion/biometric recognition tools — not to all AI-assisted content. Content edited and reviewed by a named human with editorial responsibility is exempt; small assistive edits like spellcheck or cropping don't count as AI generation, but summaries, rewrites, and composite imagery do. The generic ✨ sparkle icon is likely insufficient; the Commission published a specific official AI icon set and Code of Practice requiring clear, accessible, persistent labelling paired with plain-language text like "AI-generated." Similar mandatory labelling laws already exist in China (since September 2025), California (aligned to the same August 2, 2026 date), South Korea (since January 2026), and India (since February 2026).
Table of contents
What Actually Needs LabellingNot All AI-Generated Content Must Be LabelledThe Fine Line Between “Edited” And “AI-Generated”AI Sparkles Probably Not EnoughIt Isn’t Just EUWrapping UpMeet “Design Patterns For AI Interfaces”Useful ResourcesQuestions this post answers
When does AI content labelling become legally required under the EU AI Act?
AI labelling became a legal requirement starting August 2, 2026, under Article 50 of the EU AI Act, for any company serving EU citizens regardless of where it is headquartered. It applies to deepfakes, chatbots and AI agents, fully AI-written text on public-interest topics without human review, and emotion recognition or biometric categorization tools. Teams shipping AI features across regions can track rollout details like this on daily.dev.
Does editing AI-generated text before publishing exempt it from EU AI Act disclosure rules?
Yes, but only if the edit is substantive and a named person or entity takes editorial responsibility for it. Small assistive edits like spellcheck, grammar fixes, formatting, cropping, or color correction do not count as removing the AI-generated status; a person merely skimming the text before publishing also does not qualify as sufficient editorial review. Anyone weighing editorial workflows against AI disclosure rules can follow this kind of compliance detail on daily.dev.
Is a generic AI sparkle icon enough to comply with EU AI labelling requirements?
No, the sparkle icon is likely too ambiguous because it usually signals an "AI-powered feature" rather than that specific content was AI-generated. The European Commission published a distinct official AI icon set as part of its Code of Practice, and states that using any icon alone does not establish legal compliance; a barely visible icon, a buried footer note, or a flashing label are all explicitly non-compliant. Designers deciding how to signal AI-generated content can weigh these disclosure patterns via daily.dev.