AI Regulation Tracker / Electoral broadcast rules
Morocco's answer on AI election content: the prohibition is the rule, the labelled exception is narrow
Every election regulator is being asked whether a disclosure label is enough to cure synthetic campaign content. Article 6 of Morocco's CSCA decision 50-26 answers with a qualified no, and it repays reading even if you will never touch a Moroccan broadcast licence.
What does article 6 actually say?
Here is the operative sentence in the French text published by HACA.
Les services de communication audiovisuelle s'abstiennent de diffuser tout contenu électoral généré ou transformé par l'intelligence artificielle (IA) susceptible d'induire le public en erreur ou de porter atteinte à l'intégrité et à la crédibilité du débat électoral.
Decision CSCA 50-26, chapitre I, article 6, first sentence
My English: audiovisual communication services shall refrain from broadcasting any election content generated or transformed by AI that is liable to mislead the public or to undermine the integrity and credibility of the electoral debate.
Two words do most of the work. The first is transforme, transformed. The rule does not only reach content conjured from nothing. It reaches real footage that AI has altered, which is where the harder cases sit. The second is the disjunctive test: content is caught if it is liable to mislead the public or if it damages the integrity and credibility of the electoral debate. A broadcaster cannot escape by showing that no viewer was actually deceived.
The permission that follows is deliberately small: Ce type de contenu ne peut être diffusé qu'à des fins d'information, d'explication ou de vérification, sous réserve d'indiquer clairement la nature transformée ou artificielle du contenu concerné. My English: this type of content may be broadcast only for purposes of information, explanation or verification, provided the transformed or artificial nature of the content is clearly indicated.
Read the order of operations. Purpose first, label second. The label is a condition attached to an exception already limited by purpose, not a cure a broadcaster can apply to otherwise prohibited material.
Who does this bind, and who does it not?
This is the part most likely to be misread. Decision 50-26 is addressed to services de communication audiovisuelle, which article 1 defines as the services edited by Morocco's national public audiovisual companies and by private audiovisual operators, following law 77.03. That means licensed radio and television.
Morocco has not, through this instrument, banned election deepfakes on the internet. Nothing in the thirty articles reaches social platforms, messaging services, video sharing sites or campaign material outside a licensed broadcast. A candidate who posts a synthetic attack video to a social feed is not touched by article 6. The same video handed to a licensed channel and aired is.
That narrowness follows from who wrote the rule. HACA is a broadcast regulator acting under law 15-11 and law 77.03, and it regulates licensees. Any wider reach would need a different instrument and authority, and I am not asserting one exists.
Where exactly does the line between ban and labelled exception fall?
Article 6 does not stop there. It adds a third layer for material used pedagogically or illustratively: any sound, visual or audiovisual content generated or artificially modified must then satisfy three conditions together. My English renderings follow each French indent.
- Accompagné d'un label d'identification clair garantissant l'information du public sur sa nature: accompanied by a clear identification label ensuring the public is informed of its nature.
- Reconnaissable de manière claire et non équivoque par le public: recognisable by the public in a clear and unambiguous way.
- Non fondé sur la reproduction ou la simulation de l'apparence, de la voix ou de l'identité d'une personne réelle, à moins que cela ne soit expressément signalé de manière visible, permanente et compréhensible: not based on the reproduction or simulation of the appearance, voice or identity of a real person, unless that is expressly signalled in a visible, permanent and comprehensible manner.
The third indent is the transferable one. Most disclosure regimes treat a label as a binary: disclosed or not. Morocco separates the label from recognisability, then treats the synthetic depiction of a real person as its own category requiring a signal that is visible, permanent and comprehensible. Permanent is the operative word. A two-second disclosure card that runs and disappears does not obviously satisfy it. A persistent on-screen marker does.
Article 6 does not sit alone either. Article 5 asks broadcasters, while respecting freedom of expression, to contribute to the fight against false information liable to affect the normal conduct of the elections. Article 4 lists content that election-period programmes may not contain at all, including material that harms human dignity or privacy. A synthetic depiction of a real candidate can fail more than one of these at once.
How does this compare with other electoral AI rules on the tracker?
The useful axis is not who has a rule. It is whether disclosure alone is treated as sufficient.
| Jurisdiction and instrument | Surface regulated | Is a label alone enough? | Where the line falls |
|---|---|---|---|
| Morocco, CSCA decision 50-26, art. 6 | Licensed radio and television only | No | Prohibition first. Exception limited to information, explanation or verification, with clear indication. Real-person depictions need a visible, permanent, comprehensible signal. |
| Brazil, Resolucao TSE 23.755 | Electoral propaganda, including platforms and campaigns | No | AI-use notice mandatory for AI-made or significantly altered campaign content, but deepfakes of real people are banned outright and new synthetic content is blacked out around polling day. |
| South Korea, Public Official Election Act art. 82-8 | Campaign content generally, including online | Depends on timing | Realistic AI campaign audio, image and video is banned outright inside the 90-day pre-election window. Outside it, a clear AI-generated label is the duty. |
| Morocco, CNDP and law 09-08 | Personal data processing, any sector | Not the test | A different regulator and question. The CNDP frames AI through data protection duties, not electoral content rules. |
The Brazilian and Korean entries are the closest siblings on this tracker. Our coverage of Brazil's TSE resolution 23.755 sets out a regime pairing a mandatory AI-use notice with an outright ban on deepfakes of real people, the same two-tier structure Morocco reaches by a different route. Our coverage of South Korea's article 82-8 shows a third design, where identical content is banned or merely labelled depending on how close the election is.
Morocco already appears on this tracker through its data protection authority. Our entry on the CNDP communique on AI and personal data records that regulator's position that law 09-08 already governs AI processing. The two instruments do not overlap. One governs how personal data may be processed, the other what a licensee may air. A broadcaster running a synthetic segment can answer to both.
Why should a US professional care about a Moroccan broadcast rule?
First, if your client produces, licenses or places political content that reaches a Moroccan licensed channel between 15 August and 22 September 2026, the licensee carries the duty and will push it up the chain contractually. Agreements written before this decision lack the representation the broadcaster now needs, that supplied material is either free of AI generation and transformation or falls inside the article 6 exception with the right signalling attached. That is a contract amendment, not a policy memo.
Second, the drafting is a template. Regulators copy each other, and this text is short and easy to lift. A compliance function built around a single disclosure field will be poorly placed when the next rule asks whether the disclosure is permanent, whether the content is independently recognisable, and whether a real person is depicted.
Third, the purpose limitation is worth borrowing regardless of any legal duty. Deciding in advance that synthetic depictions of real people are allowed only for information, explanation or verification gives an editorial team a line it can apply without a lawyer.
What does this decision not do?
It does not reach online platforms, and I decline to write that Morocco has banned election deepfakes. The text does not support that.
I am also not printing a sanctions figure. Article 6 states the obligation. Consequences for breach come from HACA's supervisory powers under law 15-11 and law 77.03, which I did not open for this piece.
Nor am I calling this a first. The decision claims no novelty, and the tracker already carries earlier electoral AI rules in Brazil and South Korea. What is unusual is the drafting precision, not the timing.
One further point of care. HACA's own guide summarises article 6 more loosely than article 6 reads. Its English text covers content "fabricated or generated by artificial intelligence and is likely to mislead the public", permitted only for informing, explaining or fact-checking and "must be clearly identified as such". That drops the word transformed and omits the three cumulative conditions. Build the checklist from the decision, not the guide.
Frequently asked questions
Has Morocco banned election deepfakes across the internet?
No. Decision 50-26 is addressed to services de communication audiovisuelle, the radio and television services edited by Morocco's public broadcasting companies and by licensed private operators under law 77.03. Nothing in it reaches social platforms, messaging apps or websites. Reading it as a general internet rule is the common error.
Does labelling AI-generated election content make it lawful to broadcast in Morocco?
Not on its own. Article 6 states the prohibition first, then allows such content only for information, explanation or verification, and only if its transformed or artificial nature is clearly indicated. For material used pedagogically or illustratively, article 6 adds three cumulative conditions: a clear identification label, unambiguous recognisability by the public, and no reproduction or simulation of a real person's appearance, voice or identity unless expressly signalled in a visible, permanent and comprehensible manner.
What is the difference between decision 50-26 and the guide HACA published on 28 July 2026?
They are two separate acts. The Conseil superieur de la communication audiovisuelle deliberated decision 50-26 on 16 June 2026, and article 30 provides for notification to operators and publication in the Bulletin Officiel. On 28 July 2026 HACA published a plain-language guide, in Arabic, Amazigh, French and English. The decision binds broadcasters. The guide explains it.
When do the Moroccan election-period broadcast rules start to apply?
Article 1 defines the periode electorale as thirty-nine days, from the first hour of Saturday 15 August 2026 to midnight on Tuesday 22 September 2026, with the official campaign period fixed at thirteen days from Thursday 10 September 2026. The general legislative elections are held on 23 September 2026.
Last verified: July 28, 2026