Direct answer: From 2 August 2026, the EU AI Act’s Article 50 transparency rules apply and enforcement begins. Businesses should disclose when people interact directly with certain AI systems, label covered deepfakes and unreviewed AI-generated public-interest text, and ensure providers add machine-readable marks to covered synthetic content. The EU’s July 2026 AI Omnibus moved the main high-risk-system deadlines to 2 December 2027 for specified sensitive uses and 2 August 2028 for AI embedded in regulated products—so do not confuse those later dates with the transparency deadline.
- List every public-facing chatbot, AI agent, generator and AI-assisted publishing workflow.
- Write down whether you are the provider (you place the system on the market under your name) or deployer (you use it professionally).
- Add a clear “You are interacting with AI” notice where Article 50 applies.
- Check that generated image, audio, video and text outputs carry detectable machine-readable marking where required.
- Add visible labels to covered deepfakes and unreviewed AI-generated public-interest text.
- Record the owner, disclosure wording, marking method, review date and evidence for each system.
- Escalate uncertain or high-risk uses to qualified EU legal/compliance counsel.
This is a practical explainer, not legal advice. Scope and exceptions depend on the system, content and your role.
What changes on 2 August 2026?
According to the European Commission’s 31 July announcement, the AI Office and national authorities begin enforcing the AI Act on 2 August 2026. On the same date, Article 50 transparency requirements start applying.
| Situation | Practical requirement | Usually responsible |
|---|---|---|
| A person interacts directly with an AI system | Inform the person that they are interacting with AI, unless this is obvious from the circumstances and context | Provider |
| AI generates or substantially manipulates text, image, audio or video | Enable reliable detection through machine-readable marking where the rule applies | Provider |
| People are exposed to emotion recognition or biometric categorisation | Provide the required notice | Deployer |
| A deepfake is published or displayed | Clearly disclose its artificial or manipulated nature, subject to the Act’s conditions and exceptions | Deployer |
| AI text informs the public on a matter of public interest without human review or editorial control | Add a clear and perceivable disclosure | Deployer |
The Commission’s Article 50 guidelines explain scope, stakeholder roles, definitions, exceptions and examples. The official Quick Facts page is the fastest reference for a team review.
Copy-and-paste AI transparency register
Create one row per AI system or publishing workflow:
System/workflow name: Business owner: Vendor/model: Countries and audiences served: Our role: Provider / Deployer / Importer / Distributor / Unsure Direct AI interaction: Yes / No Synthetic content types: Text / Image / Audio / Video / None Deepfake use: Yes / No Public-interest text: Yes / No Human review or editorial control: Yes / No Emotion recognition/biometric categorisation: Yes / No User-facing disclosure text and location: Machine-readable marking method: Exception relied on (if any): Evidence saved: Legal/compliance reviewer: Last review date: Next review date:
Suggested disclosure wording
These are plain-language starting points, not official mandatory formulas:
- Chatbot: “You are chatting with an AI assistant. Its responses may contain errors; verify important information.”
- AI-generated image: “AI-generated image.”
- Materially altered media: “This image/video/audio was generated or materially altered using AI.”
- Public-interest text without human editorial review: “This text was generated by AI and was not reviewed by a human editor.”
A notice should be clear, timely and visible to the affected person. Do not use these samples as a substitute for checking the Commission guidelines and the facts of your use case.
Important: the high-risk deadlines changed
The Commission’s updated AI Act overview says the July 2026 AI Omnibus set a new implementation timeline:
- 2 December 2027: rules for systems in specified high-risk areas, including biometrics, critical infrastructure, education, employment, migration, asylum and border control.
- 2 August 2028: rules for high-risk AI systems integrated into regulated products such as lifts or toys.
Those extensions do not cancel Article 50’s 2 August 2026 transparency start. Providers of general-purpose AI models also face Commission enforcement from 2 August 2026 for obligations that became applicable in August 2025.
What about old models and old deepfakes?
The Commission’s Quick Facts page notes two transition details:
- A grace period until December 2026 applies to the marking obligation for generative AI systems placed on the market before 2 August 2026.
- Deepfakes generated before 2 August 2026 do not require retroactive labelling under this rule, although the Commission encourages disclosure.
Check the final legal text and official guidance before relying on a transition or exception.
AI literacy: add evidence, not just a policy
The Commission’s AI literacy Q&A recommends considering what AI is used, whether the organisation is a provider or deployer, the risks involved and the training appropriate to each group. It says no certificate or single mandatory course is required; organisations can keep internal records of training and guidance initiatives.
Save attendance, training material, role-specific guidance, risk examples and review dates. A one-page “use AI responsibly” policy without evidence of implementation is a weak control.
Penalties and enforcement
The Commission says Article 50 enforcement can involve national market-surveillance authorities, the AI Office for systems under its supervision, and the European Data Protection Supervisor for EU institutions. Its Quick Facts page lists fines of up to €15 million or 3% of total worldwide annual turnover for companies, with proportionality for SMEs and small mid-cap companies. The actual amount depends on the legal provision and facts of the case.
FAQ
Does every business using ChatGPT need to label all of its content?
No. The rules are use-case and role specific. The obligations differ for providers and deployers, and Article 50 includes scope conditions and exceptions. Human-reviewed editorial content is treated differently from unreviewed AI-generated text published to inform the public on matters of public interest.
Do ordinary spelling and editing tools trigger machine-readable marking?
The Commission says the provider marking obligation has an exception where an AI system performs an assistive function for standard editing or does not substantially alter the input data or its semantics. Apply that exception carefully to the actual workflow.
Are high-risk AI rules still due on 2 August 2026?
Not the main requirements covered by the July 2026 AI Omnibus timeline. The Commission now lists 2 December 2027 for specified sensitive high-risk uses and 2 August 2028 for AI embedded in regulated products. Other AI Act rules, including Article 50 transparency and enforcement provisions, still start in August 2026.
Is the transparency code mandatory?
The Commission describes the Code of Practice on Transparency of AI-generated Content as voluntary. Organisations that do not use it must be able to demonstrate compliance through alternative, equivalently adequate means where relevant.
Official sources
- European Commission: enforcement and transparency announcement, 31 July 2026
- European Commission: Guidelines on Transparency of AI-Generated Content
- European Commission: Quick Facts on Article 50 transparency rules
- European Commission: AI Act overview and updated implementation timeline
- EU AI Act Service Desk: Article 50 text