Skip to content
Atesto
Create account
All articles

AI laws

EU publishes final AI transparency guidelines for Article 50

The European Commission has formally published its 43 page guidelines on AI transparency, detailing rules for chatbots, deepfakes and synthetic media.

On November 12, 2026, the European Commission published its practical guidelines on Article 50 of the Artificial Intelligence Act in the Official Journal of the European Union. The 43 page communication details how organizations must handle transparency for chatbots, synthetic content, emotion recognition and deepfakes. The text provides the operational framework for rules that carry penalties of up to three percent of global annual turnover. It matters to executives, compliance officers and technology leaders because it defines exactly when a company must disclose its use of artificial intelligence to the public.

Who it applies to

The guidelines divide obligations between providers who build artificial intelligence systems and deployers who use them under their own authority. A single organization can hold both roles if it develops an interactive system internally and puts it into service under its own name.

Geography does not exclude a company from these rules. Deployers established outside the European Union fall within the scope (opens in a new tab) of the requirements if they foresee the dissemination of their artificial intelligence outputs within the Union. A foreign company generating a synthetic media advertisement targeted at European citizens must comply with the transparency duties.

In the advertising and media supply chain, responsibility rests on control. A company that commissions an agency to produce an advertisement is not a deployer if it does not make decisions about how the agency uses artificial intelligence. The agency producing the work holds the deployer obligations. However, if the client selects the specific tool or writes the prompts, the client becomes the deployer. Employees acting under corporate instructions, such as web designers or content creators, are not individual deployers. The company itself holds that legal responsibility.

Distribution intermediaries like hosting services and online platforms are generally excluded from the deployer definition when they merely transmit content created by others. This position changes if a platform uses artificial intelligence for its own professional marketing or operations, which brings the platform into scope as a deployer.

Scientific research maintains a specific boundary. Systems and outputs developed and put into service solely for scientific research (opens in a new tab) and development fall under an exclusion. If those same systems are later used for other purposes, the transparency obligations apply. Real world testing falls outside the separate premarket research exclusion.

What it requires

The communication outlines compliance requirements across four distinct domains.

First, providers of systems that interact directly with people must design their interfaces so users know they are dealing with artificial intelligence. This covers customer support bots, coding agents and social network avatars. Artificial intelligence agents that book appointments or negotiate must disclose their artificial nature and identify the person they act for. Hidden notices buried in terms of service do not satisfy this requirement.

Second, providers of systems that generate synthetic audio, video, text or images must mark those outputs in a format that machines can read. They must also provide the means to detect those marks. Fulfilling only the marking element is insufficient. The detection must ultimately rely on publicly available industry standard solutions.

Third, deployers of emotion recognition and biometric categorization systems must provide explicit notifications to the people exposed to them.

Fourth, deployers must label deepfakes and synthetic text published on matters of public interest. The guidelines define a deepfake objectively, without requiring an intent to deceive. A simulated subject must resemble something that exists or could plausibly exist. A realistic synthetic human qualifies as a deepfake, while a flying dragon does not.

For deepfakes, deployers cannot rely on the hidden technical watermark embedded by the provider. The label must be visible or audible to natural persons at the point of first exposure. Scrolling past a manipulated video on social media counts as first exposure, meaning a label placed only at the end of a video fails the test.

The rules for text generation focus on public interest content. Artificial intelligence manipulated text in corporate product descriptions or standard advertising copy sits outside this specific labeling rule, unless it makes health, safety or sustainability claims. However, synthetic text published to inform the public on matters of public interest requires a label.

Organizations can avoid labeling public interest text only if the content undergoes substantive human review and a designated person holds editorial responsibility. The guidelines describe this review as a "deliberate examination of the substance" by qualified personnel. Fact checking is a minimum requirement. Superficial grammar checks or automated reviews do not qualify. If an artificial intelligence tool makes a substantive intervention after the human review, the exception becomes void.

The dates

The transparency requirements under Article 50 of the Artificial Intelligence Act began to apply on August 2, 2026.

The European Commission formally published the practical guidelines in the Official Journal of the European Union on November 12, 2026, under the reference C/2026/6005.

While the guidelines themselves are nonbinding, they represent the official interpretation of the European Commission. The ultimate authoritative interpretation of the law remains with the Court of Justice of the European Union.

What to check now

  • Inventory all customer facing chatbots and verify they state their artificial nature during the first interaction.
  • Review contracts with marketing agencies to clarify who controls the artificial intelligence tools and who holds the deployer obligations.
  • Ensure any synthetic video or image that resembles reality includes a visible label for the audience, rather than just a technical watermark.
  • Establish a documented human review process for any synthetic text published on matters of public interest.
  • Confirm that the identity of the person holding editorial responsibility for public interest publications is publicly accessible.

The publication of these guidelines clarifies how the European Commission expects organizations to disclose synthetic content. The text provides the framework for authorities to assess technical interfaces and editorial workflows against the transparency standards.

Sources

  1. PPC Land (opens in a new tab) · ppc.land
  2. CASRAI (opens in a new tab) · casrai.org

This article is general information, not legal advice. For your company's situation, talk to a qualified lawyer.

This article was written with AI and checked, sentence by sentence, against the sources linked above.