Synthetic media has quietly become ordinary. Marketing images, voiceovers, product copy, support replies — a growing share is machine-made. Regulators have responded with a clear demand: people should know when they are dealing with AI output, and systems should be able to detect it automatically.
What the rules require
The EU AI Act's transparency obligations target two layers. Providers of generative systems must mark their outputs in a machine-readable way that flags the content as artificially generated or manipulated. Deployers must disclose to people when they interact with AI (for example, a chatbot), and must label deep fakes and AI-generated text published to inform the public on matters of public interest. Work on a dedicated Code of Practice for marking and labelling AI-generated content has been advancing through draft stages, which signals how detailed the practical expectations will be.
Three traps to avoid
- Treating it as a UI afterthought. A visible 'made with AI' badge is not enough on its own; the obligation includes durable, machine-readable marking such as embedded metadata or watermarking.
- Assuming it is only an EU problem. Several jurisdictions are moving in the same direction, and platform policies already require disclosure. Build the capability once.
- Forgetting the overlap with other law. A deepfake of a real person can also raise data-protection, personality-rights, and — following recent amendments — criminal-law issues, including new prohibitions on non-consensual intimate imagery.
How to prepare now
Inventory every feature that generates or edits content. For each, decide how you will mark the output (metadata, watermark, provenance signals) and how you will disclose AI interaction to users. Write it into your product spec and your vendor contracts, so the obligation does not land on a single overworked team weeks before it applies. Transparency is becoming a baseline expectation — the companies that treat it as a feature, not a footnote, will earn trust while others scramble.
