Europe’s AI disclosure rules become applicable today. From August 2, Article 50 of the EU AI Act requires certain AI systems and the organizations using them to tell people when a machine is part of the interaction—and to make important categories of synthetic media identifiable.
This is not a general rule that every AI-assisted pixel or sentence needs a bright warning label. It is a set of targeted duties divided between providers, which place AI systems on the market or put them into service, and deployers, which use those systems under their authority.
What providers must build into AI systems
Providers of interactive AI systems must design them so people are informed when they are interacting with AI, unless that fact is obvious to a reasonably informed and attentive person. In practice, chatbot and conversational interfaces need a clear disclosure delivered no later than the first interaction.
Providers of systems that generate or manipulate synthetic audio, images, video, or text also have a less visible engineering obligation: outputs must be marked in a machine-readable format and detectable as artificially generated or manipulated. The techniques must be effective, interoperable, robust, and reliable as far as technically feasible.
That turns provenance into product infrastructure. Teams may need to preserve signals through export pipelines, compression, resizing, format conversion, and downstream publishing—not merely attach a badge inside the original app.
What deployers must disclose to people
Deployers have separate, audience-facing duties. Anyone exposed to an emotion-recognition or biometric-categorization system must be informed that it is operating, subject to the Act’s specific conditions and exceptions.
Deployers publishing deepfakes must visibly disclose that the image, audio, or video was artificially generated or manipulated. The rule also covers AI-generated or manipulated text published to inform the public on matters of public interest. That text disclosure does not apply when the content has undergone human review or editorial control and a person or organization holds editorial responsibility for publication.
The Act includes tailored exceptions, including for some law-enforcement uses and for evidently artistic, creative, satirical, fictional, or analogous works. Those exceptions change how the disclosure may be presented; they should not be read as permission to hide a deceptive use.
The product requirement behind the label
A visible “AI-generated” note is only the front end of compliance. A practical implementation can involve interface copy, content credentials or other machine-readable markers, asset metadata, publishing controls, audit records, and rules for what survives when content leaves the original system.
For companies serving EU users, the immediate questions are concrete: Does the product identify itself at the right moment? Are generated files detectably marked? Can a professional user add the required public disclosure? Does the workflow distinguish a deepfake from ordinary editing, or public-interest publishing from internal drafting?
The European Commission’s voluntary Code of Practice offers one route for providers and deployers to demonstrate compliance with the synthetic-content obligations. Organizations that do not sign it still have to demonstrate compliance through other means.
Not every AI Act deadline lands today
Article 50’s transparency obligations apply from August 2, 2026. A limited grace period extends the machine-readable marking duty to December 2, 2026 for relevant systems placed on the market before today; previously generated content does not need to be labelled retroactively.
The more extensive requirements for high-risk systems follow later. Rules for systems used in areas such as biometrics, critical infrastructure, education, employment, migration, asylum, and border control apply from December 2, 2027. High-risk AI embedded in regulated physical products, such as machinery, toys, and lifts, follows on August 2, 2028.
Why this matters beyond Europe
Disclosure features are difficult to bolt on at the last minute or isolate neatly by geography. Once a product needs reliable marking, export-safe provenance, and publishing disclosures for European users, a global implementation may be simpler and more trustworthy than maintaining separate experiences.
For independent product teams such as SunMarc App Labs, the useful lesson is to treat transparency as part of the system design. Clear interaction notices, durable media provenance, human editorial checkpoints, and honest product copy can reduce compliance risk while making an AI feature easier to understand.
Europe has turned AI labelling from a trust signal into an operating requirement. The strongest products will make that transparency feel native—not like a legal sticker added after the experience was finished.
Relevant links
- European Commission: Guidelines on Article 50 transparency obligations
- European Commission: Navigating the AI Act
- European Commission: Code of Practice on marking and labelling AI-generated content
- EUR-Lex: Regulation (EU) 2024/1689
- SunMarc archive: AI accountability moves into the evidence era
- SunMarc archive: Europe’s AI infrastructure push