Since 2 August 2026, a student who posts a doctored image is not in the same legal position as the association that shares it. The European Commission has detailed in its guidelines and FAQ (last updated 24 July 2026) how Article 50 of the AI Act applies. The topic is rarely covered from the viewpoint of people who produce content without being media outlets: students, associations, campuses. That is where the rule will be tested.
AI content labelling: what Article 50 says
Article 50 applies from 2 August 2026. It separates two kinds of actors. Providers (who develop a system and place it on the market) must tell people they are talking to an AI, unless it is obvious, and mark generated content in a machine-readable format. Deployers (who use a system under their authority) must inform the public when they publish deepfakes or certain AI-generated texts.
There is a narrow grace period: systems already on the market before 2 August 2026 have until 2 December 2026 for machine-readable marking only. Content created before 2 August does not have to be labelled retroactively, although the Commission encourages it where possible.
Student or association: who is a deployer?
This is the core question. According to the Commission, a natural person using AI in a personal capacity, for example to create a deepfake and share it on social media, is carrying out a personal activity excluded from the AI Act. If they regularly gain an economic benefit, or act in a professional or freelance capacity, they become a deployer.
For a legal person, the Commission specifies that employees acting under its instructions are not separate deployers: the organisation is. The texts we read do not mention student associations explicitly. Our reading, which is not legal advice: an association communicating under its own name with AI tools falls on the deployer side more readily than a student posting in their own name.
Deepfakes and public-interest text: two separate duties
A deepfake is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places or events and would falsely appear authentic. Three cumulative criteria apply: resemblance, an existing subject, a deceptive appearance. The deployer must disclose it at the latest on first exposure, clearly and perceptibly, with no technical tool needed. The invisible mark added by the provider is not enough.
For evidently artistic, satirical or fictional works, the duty is lighter: an appropriate disclosure that does not hamper enjoyment of the work.
For text, labelling targets content that is published, meant to inform the public and on a matter of public interest: politics, public administration, public health, the environment, publicly debated scientific or cultural developments. Text that has undergone human review or editorial control is exempt.
Superficial, solely formal, or procedural checks (e.g. spell-checking or grammatical correction) are not considered to be human review or editorial control.
The nuance matters: proofreading for typos is not enough. It takes a review of the substance by competent people, or control by an editorial entity able to approve, alter or reject the text, including fact-checking and source checks. Someone must hold ultimate legal responsibility.
Who wins, who loses
- Winners: student media and publications that check their sources, since editorial work exempts their texts from labels; readers, who will know better what is synthetic.
- Losers: accounts mass-publishing unchecked generated text or images, more exposed once they act in a professional capacity.
- Uncertain: associations, whose deployer status will depend on national authorities' assessment.
Who enforces, and what penalties?
Enforcement mainly falls to national market surveillance authorities. The AI Office acts only in limited cases, for instance for systems built into a very large platform designated under the DSA. Fines can reach EUR 15 million or 3% of worldwide turnover, with proportionality for SMEs.
What it means for students
Three simple habits. First, announce synthetic visuals that imitate reality in any public communication. Second, truly review the substance of public-interest texts produced with AI, and take responsibility for them. Third, remember that purely personal use stays out of scope, which does not exempt it from criminal law or image rights.
This follows our analyses of what really applies since 2 August 2026, of the AI Act and local authorities and of AI agents out of control. Worth watching: the draft tech objectives the Socialists and Democrats group is preparing in the European Parliament against harmful deepfakes, reported by Politico on 20 October 2026, are still only a draft.
Sources
