Tomorrow, the transparency obligations under Article 50 of the EU AI Act (Regulation (EU) 2024/1689) become enforceable. From 2 August 2026, any organisation whose AI systems interact with EU users, generate synthetic media, or manipulate images, audio or video must clearly disclose that fact, or face fines of up to €15 million or 3% of global annual turnover, whichever is higher.
For enterprises that have spent the last two years treating "AI Act compliance" as a distant milestone, that distance has just run out. Here is what actually changes from this week, and what still doesn't.
What Content Must Be Labelled
Article 50 sets out four distinct disclosure duties, and it's worth separating them because they land on different teams:
- Interactive AI systems: Providers of chatbots and conversational agents must make clear that a person is interacting with AI, not a human, unless this is obvious from the context.
- Synthetic content: Providers of systems that generate synthetic audio, image, video or text must mark the output in a machine-readable format, detectable as artificially generated or manipulated.
- Emotion recognition and biometric categorisation: Deployers must inform any individual exposed to these systems.
- Deepfakes and manipulated public-interest content: Deployers must disclose that image, audio or video content has been artificially generated or manipulated, including AI-assisted news articles on matters of public interest.
Critically, disclosures buried in terms and conditions, or a vague "AI-powered" badge somewhere in a footer, will not satisfy the requirement. Guidance is explicit that labelling must be "clear and distinguishable" and delivered "at the latest at the first interaction or exposure."
Who Is In Scope
The extraterritorial reach here catches many organisations by surprise. The obligation applies to:
- Any provider placing an AI system on the EU market, regardless of where the company is headquartered.
- Third-country providers whose system outputs are used within the EU.
- UK and US businesses serving EU customers, even without an EU legal entity.
If your marketing team uses a generative image tool to produce assets seen by EU audiences, or your support desk runs an AI chatbot answering EU customers, you are almost certainly in scope.
The One Piece of Breathing Room
There is a partial reprieve. Under the AI Omnibus package, providers of existing generative AI systems placed on the market before the deadline have until 2 December 2026 to retrofit machine-readable marking into their outputs. New systems and new deployments from this week, however, must comply immediately. Enforcement bodies have signalled they will treat the extension as a technical grace period, not a licence to delay disclosure practices altogether.
Practical Steps for Compliance Teams
1. Audit Every Generative Touchpoint
Map every system in your organisation that produces synthetic audio, image, video or text, and every chatbot or voice assistant that touches an EU user. Marketing, HR, customer support and product teams often deploy generative tools independently of IT, which is exactly where compliance gaps hide.
2. Move to Machine-Readable Provenance
A visible watermark in the corner of an image is not sufficient on its own. Regulators expect machine-readable signals, embedded metadata that downstream platforms and detection tools can read automatically, in line with the C2PA content credentials standard. Tools like Google SynthID and Adobe's Content Credentials already build this into their generation and editing pipelines.
3. Review Vendor Contracts
If you rely on a third-party model provider for image, video or voice generation, confirm contractually that their outputs carry the required machine-readable disclosure. Liability for a missing label does not automatically transfer to your vendor; as the deployer facing EU users, you may carry the exposure.
Conclusion: Compliance Is Now Operational, Not Aspirational
Article 50 was always going to arrive. What changes on 2 August 2026 is that "we're working on it" stops being an acceptable answer to a regulator, a customer, or a journalist asking whether your AI-generated content is labelled. Enterprises that treated provenance tooling as a nice-to-have over the past year now have a hard compliance deadline attached to it.