New European Union transparency duties for artificial intelligence took effect Aug. 2, 2026, requiring machine-readable marking or visible disclosure in defined situations. The rules reach companies that provide generative systems as well as organizations that deploy them to create deepfakes or certain public-interest text. They do not require an “AI-made” sticker on every file touched by software.
The obligations come from Article 50 of the EU AI Act and split responsibility across the technology chain. A model provider may need to make synthetic output detectable, while the company publishing a deepfake may need a clear label that people can see. The content, purpose and degree of human review determine which duty applies.
Article 50 became applicable on Aug. 2, 2026
The European Commission published final guidance in July to explain the scope before the start date. Its announcement says providers and deployers face transparency obligations beginning Aug. 2, 2026, covering direct AI interactions, synthetic-content marking, deepfakes, emotion recognition and biometric categorization.
The rules apply within the EU legal framework even when a provider is based elsewhere but places a covered system on the European market. Companies therefore need to examine product availability and use cases, not only headquarters. National market-surveillance authorities and the EU AI Office have enforcement roles depending on the system.
Providers must build machine-readable marks into output
Providers of systems that generate or manipulate image, audio, video or text must design outputs so synthetic origin can be detected in a machine-readable format. The requirement aims to support technical identification across distribution channels, where an ordinary visible label may be removed or separated from the file.
The Commission’s quick-facts guide describes machine-readable marking for synthetic content and notes exceptions for assistive standard editing or changes that do not substantially alter the input or its meaning. A basic correction tool therefore does not automatically trigger the same treatment as a system generating a realistic new scene.
Deployers must visibly disclose deepfakes and some text
An organization using AI to generate or manipulate an image, audio recording or video that falsely appears authentic must disclose that the content is artificial. The visible notice is aimed at the audience encountering the material, not only a downstream detection tool. Artistic, satirical and similar works can receive a disclosure adapted to avoid spoiling the experience while still meeting the law.
AI-generated or manipulated text published to inform the public on a matter of public interest also requires disclosure when it lacks human review or editorial control. That condition matters for newsrooms and corporate communications: meaningful human review can change whether the specific text-labeling duty applies, though other consumer or professional rules may still require transparency.
Chatbots and emotion systems carry separate notices
Providers of interactive AI must inform people that they are interacting with a machine unless the artificial nature is obvious to a reasonably informed and attentive person. Deployers of emotion-recognition or biometric-categorization systems must notify exposed individuals, subject to the Act’s conditions and other data-protection law.
These duties show why “AI content labeling” is only shorthand. Some notices concern the identity of a conversational system, others concern the origin of media, and still others concern analysis performed on a person. A single watermark policy cannot necessarily satisfy every category.
The voluntary code offers a compliance route
The Commission published a Code of Practice on marking and labeling AI-generated content in June 2026. The code is voluntary but supplies practical measures that signatories can use to demonstrate compliance. The EU also released optional icons for disclosures.
Voluntary does not mean the underlying legal duty is optional. A company can choose another compliant method, while regulators can assess whether its approach actually meets Article 50. Documentation of technical design, testing and publishing procedures will matter when a company needs to explain that choice.
The Commission’s current summary lists fines for covered violations of up to €15 million or, for companies, up to 3 percent of total worldwide annual turnover, with proportional treatment for smaller businesses. The exact sanction depends on the provision, facts and enforcement process rather than following automatically from any missing label.
A grace period also applies to the marking obligation for some generative systems placed on the market before Aug. 2, 2026, extending into December 2026 under the amended framework. Preexisting deepfakes do not generally face retroactive mandatory labeling, although disclosure is encouraged.
Compliance starts with mapping who creates and who publishes
A company should identify each AI system, whether it acts as provider or deployer, what media leaves the system and whether people receive public-interest information without meaningful human oversight. Contracts should assign responsibility for embedded marks, retained metadata and visible labels instead of assuming another vendor covers them.
The new rules create a real labeling mandate, but precision remains essential. They regulate particular actors and uses, not every spreadsheet formula, photo adjustment or AI-assisted sentence. Organizations that map the full content path—from generation to final publication—are best positioned to apply the right disclosure at the right layer.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
