The EU AI Act's next deadline is 2 December 2026, and before anything else we need to correct ourselves. In our July update on the EU AI Act we wrote that 2 December brings a new Article 5 prohibition. It brings two, inserted as points (ba) and (bb) of Article 5(1), and their scoping conditions do more work than the headline does. That post has been corrected.
Two things apply from 2 December 2026. The two new prohibitions, and the machine-readable marking obligation under Article 50(2) for generative systems that were already on the market before 2 August 2026. Everything else that was deferred stays deferred.
What actually lands on 2 December 2026
The European Commission's own AI Act Service Desk states the position plainly. Providers of AI systems placed on the market before 2 August 2026 "have until 2 December 2026 to comply with marking and detection obligation under Article 50(2)". And "the new prohibitions relating to the generation or manipulation of non-consensual intimate material and child sexual abuse material will apply from 2 December 2026".
Those prohibitions were inserted by Regulation (EU) 2026/1744, the Digital Omnibus on AI, dated 8 July 2026 and published at OJ L, 2026/1744, 24.7.2026. The same regulation deferred the high-risk deadlines we wrote about in July. Note the number: at least one tracker circulates it as 2026/174, which is wrong, and the four-digit sequence is corroborated by the OJ header, the CELEX number 32026R1744 and the ELI identifier.
Who the two new prohibitions actually bind
This is where most coverage will get it wrong, in the same way most coverage got Article 50 wrong, so it is worth being precise.
Point (ba) prohibits placing on the market, putting into service or using an AI system that generates or manipulates realistic images, videos, audio or similar material of an identifiable natural person's intimate parts, or of an identifiable natural person engaged in sexually explicit activities, "without that person's freely-given, specific, informed, unambiguous and explicit consent for that generation or manipulation".
Point (bb) prohibits the equivalent for material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU, which is child sexual abuse material, "except where a 'without right' defence applies under national law". Some secondary coverage renders that exception as a ground excluding unlawfulness. The EUR-Lex text says "without right".
The scoping is in a new Article 5(1a), and it is what stops this catching every business that runs an image model.
For providers, placing such a system on the market is prohibited only where that generation is the system's intended purpose, or where the system's design, training, architecture, capabilities or user-facing functionality make such generation a reasonably foreseeable and reproducible outcome without significant technical modification, and the system lacks reasonable and adequate safeguards to reliably prevent it and to correct observed or reported misuse.
For deployers, use is prohibited only where the deployer uses the system for the purpose of generating that material.
A further paragraph, 5(1b), is drafted against point (ba) on its face. It opens "For the purposes of paragraph 1, first subparagraph, point (ba)" and provides that an AI system manipulating material "in a way that does not increase the exposure of any depicted intimate parts or alter the nature of any depicted sexually explicit activities shall not constitute manipulation". Whether that carve-out reads across to (bb) is a question for counsel, not for us.
In plain terms: this is aimed at nudifier applications and at people who deliberately use general tools to produce this material. A South African business running a general image model inside a product, with safeguards in place and a reporting route when misuse is found, is not the target. The two things that move you towards the target are the absence of safeguards and the absence of a correction process when someone tells you it happened. Both are governance questions rather than model questions.
The marking deadline you might already be past
Article 50 has applied since 2 August 2026. The Commission's Article 50 FAQ sets out the single grace period, verbatim: "A limited grace period is envisaged only for AI systems placed on the market before 2 August 2026 and only as regards the marking and detection obligation for AI-generated content (Article 50(2) of the AI Act). Providers of such systems must comply with those obligations only as from 2 December 2026."
Three things follow. The grace applies only to systems placed on the market before 2 August 2026, only to the Article 50(2) marking and detection duty, and only to providers. Content generated before 2 August 2026 does not need retroactive labelling.
If you built and shipped a generative feature before August and told yourself you would sort out machine-readable marking later, later is 2 December.
Everything else in Article 50 is already live and unchanged, including the role allocation we set out in July: paragraphs 50(1) and 50(2) bind providers, paragraphs 50(3) and 50(4) bind deployers. Ordinary marketing copy is not public-interest text, a generated stock-style image is not a deepfake, and where a third-party tool generates the content the 50(2) marking duty sits with that tool's provider rather than with you.
Separately, the Commission published a voluntary Code of Practice on Transparency of AI-Generated Content in June 2026. On 31 July 2026 the Commission reported that about 190 organisations had signed by the end of that month, with 82 signing Section 1 and 152 signing Section 2, so the two are not a split of the 190. Counts move. It is voluntary, it is not a safe harbour, and signing it is a reasonable signal rather than a compliance strategy.
The penalty numbers, and the one that applies to you
Regulation 2026/1744 did not touch Article 99. The tiers stand as enacted:
- Article 99(3), breach of the Article 5 prohibitions: up to EUR 35 million or 7% of total worldwide annual turnover, whichever is higher.
- Article 99(4), breach of other operator obligations including Article 50 transparency: up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher.
- Article 99(5), supplying incorrect or misleading information to authorities: up to EUR 7.5 million or 1%, whichever is higher.
And the provision nobody quotes, which is the most useful one in this post for our readers. Article 99(6) inverts the calculation for SMEs and start-ups: each fine is up to the percentages or the amounts in paragraphs 3, 4 and 5, whichever is lower.
For a South African business turning over R40 million, that is the difference between a theoretical exposure of EUR 15 million and a percentage of turnover. Note also that the new prohibitions sit in the Article 5 tier, which is the expensive one, and Article 50 transparency breaches sit in the 3% tier. The widely quoted 35 million figure is routinely attached to transparency obligations in secondary coverage. It does not belong there.
What has not moved
The deferrals from July hold. High-risk obligations under Chapter III apply from 2 December 2027 for systems classified as high-risk under Article 6(2) and Annex III, and from 2 August 2028 for those classified under Article 6(1) and Annex I, meaning AI embedded into regulated products. General-purpose AI obligations applied from 2 August 2025 and became enforceable on 2 August 2026.
A sourcing note, because it belongs next to the dates rather than at the end. Recital 40 of the amending regulation states both deferred dates explicitly, and that is primary text. What we could not read is the operative amending provision for Article 113 itself, because EUR-Lex truncates before it on every route we tried. The Commission's own enforcement and regulatory framework pages give the same two dates, and we have used those as operational confirmation. So: primary text for the intent, Commission pages for the operational position, and no claim from us that "Article 113 as amended provides". If a legal opinion turns on the statutory wording, take it from the Official Journal PDF.
What to do between now and 2 December
Four things, and none of them takes six weeks unless you have left it late.
- Determine your role per system, not per company. You are probably a deployer of most things and a provider of one or two. The duties differ, and a system your developers wrapped and shipped may have made you a provider without anyone deciding to be one.
- Find anything generative you placed on the market before 2 August 2026. That is the population the marking deadline applies to. If the generation happens in a third-party tool, confirm in writing whose duty the marking is, because in most cases it is theirs.
- Write down the safeguards. For the new prohibitions, the provider test turns on whether misuse is a reasonably foreseeable and reproducible outcome without adequate safeguards, and whether you correct reported misuse. A safeguard nobody documented is hard to point at later.
- Put a reporting route in place. The correction limb of the test assumes someone can tell you, and that something happens when they do.
If none of this applies because you have no EU market exposure at all, you still have POPIA, and we set out what that means in AI regulation in South Africa. The domestic position is less generous than people assume, mostly because there is no guidance to lean on.
If you are unsure whether it applies, that is a scoping question rather than a legal one, and it is a reasonable place to start an AI consulting conversation before it becomes a legal one.
FAQ
Does the EU AI Act apply to a South African company? It can. The Act reaches providers placing AI systems on the EU market and deployers established in the EU, and it reaches providers and deployers outside the EU where the output produced by the system is used in the EU. Whether your specific arrangement falls inside is a question for your counsel.
What exactly changes on 2 December 2026? Two new Article 5 prohibitions take effect, covering non-consensual intimate material and child sexual abuse material, and providers of generative systems placed on the market before 2 August 2026 must comply with the Article 50(2) marking and detection obligation.
We use a third-party tool to generate images. Do we have to mark the output? The Article 50(2) marking duty sits with the provider of the system that generates the content. Where that is a third-party tool, it is their duty. Your duties as a deployer are the ones in Article 50(3) and 50(4), which cover emotion recognition, biometric categorisation, deepfakes and AI-generated text published to inform the public on matters of public interest.
Is our image feature caught by the new prohibitions? Only if generating that material is its intended purpose, or if such generation is a reasonably foreseeable and reproducible outcome without significant technical modification and you lack adequate safeguards and a correction process. A general-purpose tool with working safeguards is not the target.
What is the maximum fine for an SME? Article 99(6) inverts the usual calculation for SMEs and start-ups: the fine is capped at whichever of the fixed amount or the turnover percentage is lower, rather than higher. That is a materially different exposure from the headline numbers, and it is rarely mentioned.
Have the high-risk deadlines moved again? No. Annex III high-risk obligations apply from 2 December 2027, and high-risk systems embedded in regulated products from 2 August 2028, as amended by Regulation (EU) 2026/1744.
Is the Code of Practice on Transparency worth signing? It is voluntary and it is not a safe harbour. Signing signals intent and gives you a structure to work from. It does not discharge the Article 50 obligations, and it should not be presented internally as though it does.
References
- Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI), OJ L, 2026/1744, 24.7.2026, CELEX 32026R1744. Article 5(1) points (ba) and (bb) and Article 5(1a) and (1b) are quoted from this text; recital 40 is the source for the deferred high-risk dates. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202601744
- Consolidated Regulation (EU) 2024/1689 as amended, CELEX 02024R1689-20260727. https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX%3A02024R1689-20260727
- European Commission, AI Act Service Desk, "When does enforcement start?" (accessed September 2026). https://ai-act-service-desk.ec.europa.eu/en/ai-act/faq/when-does-enforcement-start
- European Commission, "Transparency obligations under Article 50 of the AI Act", FAQ, last updated 24 July 2026. https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act
- European Commission, "Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act", adopted 20 July 2026. https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems
- European Commission, "Code of Practice on Transparency of AI-Generated Content", published 10 June 2026. https://digital-strategy.ec.europa.eu/en/faqs/code-practice-transparency-ai-generated-content
- European Commission, "Strong backing for the Code of Practice on Transparency of AI-Generated Content", published 31 July 2026, reporting about 190 signatory organisations by the end of July, 82 to Section 1 and 152 to Section 2. https://digital-strategy.ec.europa.eu/en/news/strong-backing-code-practice-transparency-ai-generated-content
- European Commission, "AI Act enforcement framework" (last updated 24 August 2026). https://digital-strategy.ec.europa.eu/en/policies/enforcement-ai-act
- AI Act Service Desk, Article 99 (penalties). https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-99
- Directive 2011/93/EU on combating the sexual abuse and sexual exploitation of children and child pornography. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32011L0093
This post is not legal advice. Three questions need your own counsel: whether your EU market exposure brings you inside the Act's territorial scope at all; which role, provider or deployer, you occupy for each system you run; and whether a specific generative feature falls inside the new Article 5(1) points (ba) or (bb) given its intended purpose and the safeguards you have in place. Interpretation of the Digital Omnibus amendments is new and will develop.
Written by JP Dippenaar, Sixees Labs. Last reviewed September 2026.