EU AI Act transparency rules: what AI agent builders should check

Reviewed September 29, 2026 against the European Commission's Article 50 FAQ, transparency guidance, and the AI Act text. This is a builder's issue-spotting guide, not legal advice. The rules and their application can change; have qualified counsel review a specific deployment.

The AI Act's Article 50 transparency obligations began applying on August 2, 2026. They are specific to certain systems and uses. A business should map each agent's actual function, who provides it, who deploys it, where it is used, and what a person sees before deciding what it must disclose. The Commission's current FAQ explains scope and exceptions.

Start with the interaction

Article 50(1) concerns providers of AI systems intended to interact directly with natural persons. Those systems must be designed so people are informed they are interacting with AI, unless that is obvious in context. If your customer support agent chats directly with people, review the notice in the interface and confirm who is responsible for providing and deploying the system. The duty does not automatically cover every email that an AI assistant helped a human draft.

Treat generated content as a separate question

Article 50 also contains duties for providers to mark certain AI-generated or manipulated audio, image, video, and text outputs in a machine-readable way, subject to the text's conditions and exceptions. Deployers have separate disclosure duties for deepfakes and for certain AI-generated or manipulated text published to inform the public on matters of public interest. Human review or editorial control matters to the latter duty. A CMS flag alone is not proof of compliance. Read the Commission's guidance and assess the relevant system and output.

The Commission says a limited transition until December 2, 2026 applies to the Article 50(2) marking and detection obligation for systems placed on the market before August 2, 2026. It does not create a general grace period for Article 50. Commission FAQ.

A practical review for a small team

1. List each AI system, its provider and deployer, intended users, and the EU contexts in which it operates.
2. Identify any direct interaction with people and test whether the AI nature is clear at the point of use.
3. Record what content the system generates or alters, where it appears, and which Article 50 provision may apply.
4. Check exceptions and transition provisions against the official text and guidance. Keep evidence of the decision and the implementation.
5. Escalate uncertain cases, especially employment, credit, education, health, or public service decisions, for legal review. Those use cases can raise obligations beyond Article 50.

This inventory and review process is my operational suggestion. It is not a statutory checklist. The AI Act has different rules for prohibited, high-risk, general-purpose, and transparency-related uses; a single agent can raise more than one issue. The Commission's AI Act overview explains the categories and application timeline.

What changed since this article was published

The original June 2026 version used a countdown to August 2 and stated that any AI-generated customer content required disclosure. That was too broad. This review replaces the countdown with the current application date and ties each duty to its actual Article 50 scope. It also corrects the fine figure: the Commission FAQ describes potential Article 50 fines of up to €15 million or 3% of worldwide annual turnover, subject to the regulation's conditions and proportionality. This is a ceiling, not a prediction of enforcement against a specific business.

For teams building marketing agents, the practical next step is to use a quality review before release and document who owns the final decision.