The European Union’s most visible rules for AI-generated content take effect on 2 August 2026. Under Article 50 of the EU AI Act, certain chatbots must identify themselves, providers of generative AI systems must make synthetic outputs detectable, and people or organisations publishing deepfakes or some AI-written public-interest material must clearly disclose that AI was involved.
This is an EU law, but its reach does not stop at Europe’s borders. Indian AI companies offering products in the EU, agencies producing campaigns for European clients, and publishers or creators whose AI outputs are used in the EU may fall within its territorial scope. The practical lesson is simple: “AI-assisted” and “AI-generated” are no longer merely editorial labels for some cross-border work. They can be compliance requirements.
What changed on 2 August 2026?
The AI Act has been arriving in stages. Prohibitions on a narrow group of unacceptable AI practices and AI-literacy duties started earlier, while rules for general-purpose AI models began applying in 2025. The transparency obligations in Article 50 apply from 2 August 2026, alongside wider enforcement of the Act.
The European Commission has also published detailed transparency guidelines and a voluntary Code of Practice. Companies do not have to sign the code, but those that do not must demonstrate compliance through other equally adequate measures.
| Situation | Main transparency duty | Who usually carries it |
|---|---|---|
| A person is directly interacting with an AI system | Inform the person that they are interacting with AI, unless this is obvious | Provider or deployer, depending on the setup |
| AI generates or materially manipulates text, audio, images or video | Make the output detectable in a machine-readable format where technically feasible | AI-system provider |
| A deepfake is published or shown | Clearly disclose that the content was artificially generated or manipulated | Person or organisation deploying the AI |
| AI-written text informs the public on a matter of public interest | Disclose that the text was AI-generated or manipulated, subject to an editorial-control exception | Publisher or deployer |
| Emotion-recognition or biometric-categorisation AI is used | Inform exposed individuals, alongside applicable data-protection duties | Deployer |
Machine-readable marking is different from a visible label
The Act separates two ideas that are often mixed together. Providers of generative AI systems must support machine-readable marking so synthetic content can be detected. Depending on the medium, this could involve provenance metadata, watermarks, fingerprints or other techniques that survive normal distribution as far as reasonably possible.
A visible disclosure is a separate responsibility in key use cases. A deepfake video may need an on-screen notice or an equally clear disclosure that an ordinary viewer can understand. Hidden metadata alone does not necessarily tell the person watching it that the scene or voice was generated or altered.
The marking duty is not absolute in every editing workflow. The law and Commission guidance recognise exceptions for standard assistive editing functions and systems that do not substantially alter the input or its meaning. Removing background noise, correcting colour or fixing spelling is not automatically treated like generating a convincing synthetic event.
Which AI-written articles need a disclosure?
Article 50 targets AI-generated or manipulated text published to inform the public on matters of public interest. That can include news, politics, public health, elections, safety, regulation or other subjects affecting public debate.
There is an important exception when AI output has gone through human review or editorial control and a person or legal entity holds editorial responsibility for publication. That does not make accuracy, copyright or platform rules disappear, but it means the AI Act is not designed to force a warning onto every professionally edited sentence that began with an AI tool.
For publishers, the safest workflow is to document who reviewed the output, what facts were checked and who approved publication. A casual glance should not be presented as meaningful editorial control. Organisations should also keep a consistent disclosure policy for cases that remain in scope.
What Indian businesses and creators should check
The EU AI Act can apply to providers placing AI systems or models on the EU market, deployers located in the EU, and some providers or deployers outside the EU when an AI system’s output is used in the EU. Simply having a website that Europeans can technically open does not by itself settle the legal analysis; the product, customer, deployment and use of the output matter.
Indian teams should map their exposure using five questions:
- Do we sell or provide an AI system to EU customers? A SaaS product, chatbot, image generator or API offered in Europe needs a territorial-scope review.
- Do we create synthetic media for an EU client? Contracts should say who adds visible disclosures, preserves provenance data and responds to complaints.
- Do we publish deepfakes or realistic synthetic voices? Entertainment and artistic works may use an appropriate disclosure that does not spoil the experience, but disclosure is still the baseline.
- Do we use AI for public-interest publishing? Record the human editorial process and decide whether a visible AI notice is required.
- Does our tool strip metadata? Resizing, exporting, messaging apps and social networks can remove provenance information. Test the entire publishing chain, not only the original file.
A practical compliance checklist
- Inventory every customer-facing chatbot, voice bot, synthetic-media tool and automated publishing workflow.
- Separate minor AI-assisted editing from generation or manipulation that changes meaning.
- Add a plain-language AI disclosure that is easy to notice before or at first exposure.
- Preserve machine-readable provenance and test whether it survives export, compression and upload.
- Assign responsibility between the tool provider, agency, client and publisher in writing.
- Keep evidence of human review for public-interest text, including the reviewer and approval date.
- Create a process for corrections, complaints and removal requests.
- Review privacy, copyright, consumer-protection and platform obligations separately; an AI label does not replace them.
What the new rules do not mean
The rules do not require every photograph touched by an AI feature to carry a dramatic warning. They do not automatically ban deepfakes, and they do not make AI-generated content truthful or safe merely because it has a label. Their purpose is transparency: people should know when they are dealing with a machine or materially synthetic content in situations where deception is a real risk.
They also do not replace technical security. Provenance marks can be removed, copied or lost, and a malicious actor may ignore disclosure duties entirely. As our explainer on AI-agent containment failures shows, trustworthy AI needs operational controls as well as policy.
What happens next?
The Commission’s AI Office and national authorities will oversee enforcement. The Commission says more than 180 organisations have already signed the transparency code, giving the market a common implementation reference. Older content created before 2 August 2026 does not generally need to be labelled retroactively, although the Commission encourages voluntary disclosure where practical. A limited grace period applies to the machine-readable marking obligation for some systems placed on the market before the deadline.
Indian organisations with EU-facing AI work should treat 2 August as the start of an operating requirement, not a one-time policy announcement. The durable approach is to build disclosure, provenance and editorial accountability into the workflow before content is generated. This article is a practical overview, not legal advice; businesses with material EU exposure should obtain advice for their specific product and distribution model.
Sources: European Commission enforcement announcement dated 31 July 2026; Commission Article 50 transparency FAQ; and the official EU AI Act text.
