Tomorrow is the deadline. From August 2, 2026, any company operating in the European Union must label AI-generated content, disclose chatbots as artificial, and mark deepfakes as manipulated. The rules come from Article 50 of the EU AI Act, the transparency provisions that everyone knew were coming but many are unprepared for.
🔍 THE BOTTOM LINE
Article 50 is the first broadly enforceable piece of the EU’s sweeping AI law, and it touches more organisations than almost any other provision. Every business with EU users — not just AI labs — faces chatbot disclosure, content watermarking, and deepfake labelling. Fines for non-compliance reach €35 million or 7% of global revenue, whichever is higher. Existing systems have until December 2 to adapt.
What Article 50 Actually Requires
The EU AI Act’s transparency rules impose four distinct obligations:
-
Chatbot disclosure. Any website, customer service department, or government agency using a chatbot must inform users they are interacting with AI — unless the context already makes it obvious. The AI provider is responsible for designing this disclosure into the system.
-
AI-generated content marking. Images, video, or text generated by AI must carry a machine-readable marker — a watermark, embedded metadata, or technical tag — that allows detection tools to identify it as AI-produced. This marker must survive cropping and basic alteration.
-
Deepfake labelling. Deepfakes — video, audio, or images that appear real but are AI-generated or manipulated — must be clearly labelled in “plain and simple natural language.” An audio deepfake mimicking a politician must begin with a spoken disclaimer stating it is AI-generated.
-
Emotion and biometric notification. Systems using emotion recognition or biometric categorisation (such as facial recognition) must inform people exposed to them.
Crucially, the machine-readable marker for AI-generated images and text does not need to be visible to the end user. It must be detectable by verification tools, but a casual viewer may not see a “made by AI” stamp on their feed. Deepfakes, however, require visible disclosure.
The Exemptions and the Grace Period
The rules affect content made for professional reasons. Individuals using AI in a “purely personal capacity” are exempt. Content created for “artistic, creative, satirical, fictional” purposes is also exempt from the deepfake labelling requirement, though the machine-readable marking still applies to the underlying AI system.
Existing AI systems — those deployed before August 2 — have until December 2, 2026 to comply. New systems must comply from day one.
Text that aims to inform the public on general interest issues must be labelled if created using AI without human editorial oversight. This is the provision that catches newsrooms using AI to draft articles without a human editor reviewing the output.
What Companies Are Already Doing
The major platforms have moved early. TikTok has required AI-generated content labels from creators for several years and says over three billion items already carry labels. Meta has deployed its “AI Info” label across Instagram and Facebook. Google has signed the EU’s code of conduct on AI transparency and is working with Nvidia, OpenAI, and Apple on digital tagging tools.
But these are the platforms with the engineering resources to implement watermarking at scale. The compliance gap is wider for smaller companies — marketing agencies using AI to generate ad copy, customer service firms deploying chatbots, or newsrooms experimenting with AI-assisted writing.
Why This Matters Beyond Europe
The EU AI Act has a Brussels Effect. Companies that build for the EU market tend to apply those standards globally rather than maintaining separate codebases. When Apple complies with EU charging rules, the world gets USB-C. When the EU demands AI labels, the infrastructure for content provenance — watermarks, C2PA certificates, detection APIs — gets built into products everywhere.
New Zealand companies using AI in customer-facing contexts should watch this closely. If your chatbot serves EU users, even incidentally, Article 50 applies. If you generate marketing content with AI and distribute it in Europe, the machine-readable marking requirement applies. The EU’s earlier designation of ChatGPT and Roblox as very large online platforms under the DSA signalled the direction. Article 50 extends the compliance net much further.
This also connects to the EU AI Act’s broader rollout. The high-risk AI provisions for employment and education also reference August 2 as a key date, though those rules have faced delays and rollback proposals. The transparency provisions in Article 50 are the first to hit with real enforcement teeth.
The Enforcement Question
Fines of up to €35 million or 7% of global revenue are the headline figure, but enforcement reality is messier. The EU has been criticised for placing more demands on businesses, and the practical implementation challenges are significant — particularly for the machine-readable marking requirement, which assumes a level of content provenance infrastructure that does not yet exist at scale.
“We have heard that it is going to be very, very difficult to implement,” Ashley Casovan of the International Association of Privacy Professionals told AFP. “But I think we often hear this with compliance requirements. And yet, the world turns and we figure these things out.”
The deeper question is whether labelling actually solves the problem. A machine-readable watermark tells a verification tool that content is AI-generated, but most consumers do not run verification tools. A visible “AI” label on a deepfake of a politician tells the viewer it is fake — but only if they notice it, read it, and trust the label. The labelling regime is necessary infrastructure, but it is not a complete solution to AI-driven disinformation.
❓ FAQ
Does this affect New Zealand companies? Yes, if they have EU users. Any company operating a chatbot, generating AI content, or deploying deepfake-capable systems that reach EU users falls under Article 50. The extraterritorial reach mirrors the GDPR model.
Do I need to label every AI-assisted email? No. The rules target content made for professional reasons, not personal use. An employee using AI to draft an internal email is exempt. A marketing agency using AI to generate client-facing ad copy is not.
What counts as a “deepfake” under the rules? Any text, image, video, or audio that appears real but has been generated or manipulated by AI. The key test is whether a viewer would reasonably believe it is authentic without disclosure.
What happens if a company doesn’t comply? Fines up to €35 million or 7% of global revenue, whichever is higher. Enforcement is handled by national AI authorities in each EU member state, not a single central body.
How is this different from the DSA requirements? The Digital Services Act covers online platform obligations around content moderation and risk assessment. Article 50 of the AI Act is specifically about AI-generated content transparency — disclosure, marking, and labelling.
🔍 THE BOTTOM LINE
The EU’s AI labelling law is the first concrete enforcement of the bloc’s AI Act, and it sets the global benchmark for content transparency. Tomorrow, the grace period for the biggest players ends. For everyone else — the small businesses, the marketing teams, the newsrooms — December 2 is the real deadline. The fine print matters less than the signal: AI-generated content without provenance markers is becoming a compliance liability, not just an ethical question.
📰 Sources
- The Local — Explained: The EU’s new rules on AI, deepfakes and chatbots
- Gulf News — Made by AI? EU tells firms to stick a label on it
- European Commission — Code of Practice on Transparency of AI-generated Content
- artificialintelligenceact.eu — Article 50 Transparency Rules
- TechTimes — EU AI Act Chatbot Disclosure And Deepfake Labeling
