Product news policy brief: EU AI transparency rules begin applying to interactive and generated content

On August 2, transparency obligations under Article 50 of the European Union AI Act began to apply. The European Commission's July guidance explains that people should be able to recognise when they are interacting directly with an AI system or encountering content generated or manipulated by AI. For education organizations, the practical work is to identify where AI appears in learning, communication and administration, then assign the correct notice, label or technical marking.

Providers must design relevant interactive systems to inform people that they are interacting with AI. They must also support machine-readable marking of synthetic or manipulated text, image, video and audio so its origin can be detected, subject to scope and exceptions. Deployers have disclosure duties when people encounter emotion-recognition or biometric-categorisation systems, deepfakes, or AI-generated public-interest text that has not received human review or editorial control.

These distinctions matter in schools. A student-facing support bot needs a clear AI notice. A synthetic voice or video used in public communication may need provenance and a perceivable label. A school that publishes an AI-drafted public-interest notice without meaningful human editorial control faces a different obligation from a teacher who uses AI to improve wording and then takes responsibility for the final text. Ordinary assistive editing that does not substantially alter the input or its meaning can fall outside some marking duties, depending on the detailed rules.

The Commission states that generative systems placed on the market before August 2 receive a transition for the machine-readable marking obligation until December 2026. It also describes maximum penalties of up to 15 million euros or 3% of worldwide annual turnover for companies, with proportionality provisions. Schools should obtain appropriate legal advice for their role and jurisdiction rather than treating a short product-news summary as a compliance determination.

Article 50 is not the entire AI Act and is not a complete checklist for high-risk uses in education. It does not by itself resolve whether a particular admissions, assessment, monitoring or emotion-analysis system is lawful, fair or educationally justified. Transparency also cannot make an inaccurate or discriminatory system acceptable. A visible label is one control within a larger process that includes purpose limitation, data protection, human oversight, accessibility and challenge routes.

A useful school inventory can record every interactive AI surface and every channel that publishes generated media. For each one, the institution can identify provider and deployer roles, intended audience, human review, machine-readable provenance, visible disclosure, retention, responsible owner and evidence that the notice is understandable to children. Procurement contracts should preserve the information needed to meet those duties.

For Hong Kong organizations serving EU users or partnering with European institutions, role and territorial scope need case-specific review. More broadly, the rule offers a transferable design principle: learners, families and staff should know when AI is interacting, generating or transforming content, and should know who remains accountable for the decision or message.

The most useful notice explains more than the word AI: it states the system's role, the responsible institution, the human review performed and how a person can question the result. Comprehension testing is especially important for younger users.