AI Act from 2 December 2026: New Prohibitions under Points (ba) and (bb) and Article 50(2) for Legacy Systems

Regulation (EU) 2026/1744 has added another key date to the AI Act. On 2 December 2026 two new prohibitions become applicable, and the grace period on the marking obligation ends for generative systems already on the market. This article reads both changes against the legal text: who is affected, where the boundaries run and what should be checked before the date.

In short

From 2 December 2026 two new AI Act prohibitions apply under Article 5(1), points (ba) and (bb): AI systems producing non-consensual intimate depictions of identifiable people and child sexual abuse material. On the same day, Article 111(4) ends the transitional period for machine-readable marking of synthetic content by systems placed on the market before 2 August 2026.

What applies from 2 December 2026

On Wednesday, 2 December 2026, two things change in the AI Act at once. Legally they have little in common, but they share a calendar entry. First, two new prohibitions in Article 5(1), first subparagraph, become applicable, points (ba) and (bb), together with the limiting provisions in Article 5(1a) and (1b). Second, the transitional period for machine-readable marking under Article 50(2) ends for providers of generative AI systems placed on the market before 2 August 2026. Counting from the publication date of this article, that leaves 33 days.

Both changes come from Regulation (EU) 2026/1744, the Digital Omnibus on AI. It is dated 8 July 2026, was published in the Official Journal on 24 July 2026 and, under its Article 4, entered into force on the third day after that, i.e. on 27 July 2026. The application dates themselves sit in the amended AI Act: Article 113, third paragraph, point (a) for the prohibitions, and Article 111(4) for marking by legacy systems. The wider context of the Omnibus amendments is covered in Regulation (EU) 2026/1744: AI Act deadlines 2027/28; this article concentrates on the single December date.

The addressees differ. The new prohibitions are aimed at providers that place an AI system on the market or put it into service, and at deployers that use it. The marking obligation under Article 50(2) applies only to providers of systems generating synthetic audio, image, video or text content. An organisation that merely uses generative AI is not caught by the second change in that role, but it is caught by the first as soon as it uses a system for one of the prohibited purposes.

Two changes, one date: from 2 December 2026 Article 5(1), points (ba) and (bb), and Article 5(1a) and (1b) apply (Article 113, third paragraph, point (a)), and the transitional period for Article 50(2) ends for systems placed on the market before 2 August 2026 (Article 111(4)).

The provisions at a glance

The table summarises which provision applies to whom and from when. Every entry has been checked against the consolidated text of 27 July 2026, and the dates additionally against the amending regulation itself.

RuleProvisionApplies fromAddressee
Prohibition of non-consensual intimate depictions of identifiable personsArt. 5(1), first subpara., point (ba)2 December 2026 (Art. 113, third para., point (a))Providers and deployers
Prohibition of material or performance within the meaning of Art. 2(c) and (e) of Directive 2011/93/EUArt. 5(1), first subpara., point (bb)2 December 2026 (Art. 113, third para., point (a))Providers and deployers
Limits on both prohibitionsArt. 5(1a) and (1b)2 December 2026 (Art. 113, third para., point (a))Providers and deployers
Machine-readable marking of synthetic content, systems placed on the market from 2 August 2026Art. 50(2)2 August 2026 (Art. 113, second para.)Providers
Machine-readable marking of synthetic content, systems placed on the market before 2 August 2026Art. 50(2) in conjunction with Art. 111(4)by 2 December 2026 at the latestProviders
Information on direct interaction, on emotion recognition and biometric categorisation, deep fake disclosureArt. 50(1), (3) and (4)2 August 2026, no transitional periodProviders (para. 1), deployers (paras. 3 and 4)

Point (ba): non-consensual intimate depictions

Article 5(1), first subparagraph, point (ba) prohibits “the placing on the market, the putting into service or the use of 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”, absent that person’s consent. The consent must be “freely-given, specific, informed, unambiguous and explicit” and must relate to that very generation or manipulation.

Recital 11 of Regulation 2026/1744 names the trigger plainly: the proliferation of technologies “often described as ‘nudification’ applications”. The legislator classifies non-consensual intimate material as sexual violence and abuse and grounds the prohibition in the risk to the human dignity, personal autonomy, integrity and private life of the people affected.

Four elements define the boundary. The material must be realistic; according to recital 12, this means a credible, real-life depiction of a person’s face, voice or body, whereas cartoonish or physically impossible depictions are excluded. The person must be identifiable. The material must show intimate parts or sexually explicit activity. And the depicted person must not have consented. If any one of these elements is missing, point (ba) does not apply. Criminal prosecution and national remedies protecting image, privacy and dignity are unaffected in any case, as recitals 15 and 16 confirm.

Recital 12 also sets out what falls outside the prohibition: material that does not depict identifiable persons, realistic partially nude depictions where intimate parts are not revealed and sexually explicit activities are not depicted, non-realistic artistic nude works, and generative applications where intimate parts are shown only with the consent of the depicted person. The recital gives try-on applications and medical applications, such as anatomical simulations and mammograms, as examples.

Article 5(1b) narrows the concept of manipulation. An edit that does not increase the exposure of any depicted intimate parts and does not alter the nature of any depicted sexually explicit activities does not constitute manipulation for the purposes of point (ba). Recital 12 cites changing the background, adding a text heading and enhancing contrast or brightness as examples. Conversely, any edit that increases the level of exposure or alters the nature of the activity falls within the prohibition, even where the source material was already intimate.

Point (bb): child sexual abuse material

Point (bb) prohibits the placing on the market, putting into service or use of an AI system that “generates or manipulates material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU”. Article 2(c) of the Directive defines child pornography; it covers, among other things, material visually depicting a child engaged in real or simulated sexually explicit conduct and, in point (iv), explicitly also realistic images. Article 2(e) defines a pornographic performance as a live exhibition aimed at an audience, “including by means of information and communication technology”.

Unlike point (ba), point (bb) does not recognise consent as an exception. The only carve-out is where “a ‘without right’ defence applies under national law”. Recital 13 names two groups of cases: activities carried out under domestic legal powers, such as the work of authorities in criminal proceedings, and red-teaming and evaluation activities that assess whether a system complies with the prohibition.

For organisations offering generative image, video or audio models, the second group matters in practice. Under recital 13, testing one’s own safeguards is not caught by the prohibition as such. Whether a specific test is covered by such a defence, however, depends on national law; that assessment belongs before the test starts and with the legal function, not with the engineering team alone.

When providers and deployers infringe: Article 5(1a)

The real line-drawing happens in Article 5(1a). It defines when a provider or deployer infringes point (ba) or (bb), and so prevents every generative image model from being prohibited purely because of its technical capabilities. Recital 12 says so explicitly: the prohibition should not prevent providers from developing the technical capabilities of AI systems to generate or manipulate images, videos, audio or similar material.

For providers, placing on the market or putting into service is prohibited in two cases only. In the first, generating or manipulating such material is the intended purpose of the system, which under Article 3(12) means the use intended by the provider, including what is stated in promotional or sales materials. In the second, the system’s design, training, architecture, capabilities or user-facing functionalities make that outcome “a reasonably foreseeable and reproducible outcome, without requiring significant technical modification”, and the system “does not have reasonable and adequate technical safety measures and other safeguards” to reliably prevent it and to correct observed or reported misuse.

The second case is the one that matters in practice, because it reaches general-purpose models intended for something else entirely. Recital 12 lists measures that may serve as safeguards: data cleaning, refusal training, safe prompt design and output controls, runtime prompt guardrails, content classification and filtering mechanisms, usage restrictions, abuse detection mechanisms, and notice and action mechanisms. According to the same recital, they are adequate if they align with state-of-the-art measures and demonstrably prevent or sufficiently reduce the likelihood of such outcomes in each specific case, taking into account reasonably foreseeable circumvention without significant technical modification.

For providers retaining effective control over a system, for instance through a platform or a web interface, recital 12 additionally mentions methods for following and reporting misuse cases. Where a system’s intended purpose is precisely the generation of intimate depictions, the safeguards should include means of reliably collecting and demonstrating the consent of the depicted person. Where safeguards are circumvented, adequate corrective measures should follow; the recital also points to the release and distribution strategy, such as open-source releases.

For deployers, use is prohibited under Article 5(1a), point (b), only where the deployer uses the system for the purpose of generating or manipulating such material. According to recital 12, this includes circumventing safeguards and misusing lawful systems for that purpose, but not use for lawful purposes and not accidental generation. Under Article 3(4), a deployer is anyone using an AI system under its authority, except where the system is used in the course of a personal non-professional activity; Article 2(10) additionally excludes the obligations of such private deployers. For organisations using generative systems in day-to-day work, the deployer role is the relevant one.

A further boundary concerns development and testing. Under Article 2(8), the AI Act does not apply to research, testing or development activity before a system is placed on the market or put into service; testing in real-world conditions is expressly not covered by that exclusion. Anyone trialling safeguards under real-world conditions therefore cannot rely on the research exemption.

Technical capability alone is not enough. A provider infringes point (ba) or (bb) if the prohibited outcome is the intended purpose, or occurs foreseeably and reproducibly without reasonable and adequate safeguards. A deployer infringes if it uses the system for that purpose.

Article 50(2): the marking obligation for legacy systems

Article 50(2) requires providers of AI systems, “including general-purpose AI systems, generating synthetic audio, image, video or text content”, to ensure that the outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated”. The technical solutions must be effective, interoperable, robust and reliable as far as technically feasible, taking into account the specificities and limitations of the various types of content, the costs of implementation and the generally acknowledged state of the art, as may be reflected in relevant technical standards.

The paragraph itself sets out three exceptions: systems performing an assistive function for standard editing, systems that do not substantially alter the input data provided by the deployer or its semantics, and systems authorised by law for criminal enforcement purposes. The Regulation does not define standard editing. Anyone relying on one of these exceptions should record the reasoning for each system.

In principle, Article 50 has applied since 2 August 2026, the general date of application under Article 113, second paragraph. For systems already on the market, Regulation 2026/1744 inserted an exception: under the new Article 111(4), providers of systems placed on the market before 2 August 2026 “shall take the necessary steps in order to comply with Article 50(2) by 2 December 2026”. Recital 38 explains this four-month transitional period as giving providers reasonable time to adapt their practices without disrupting the market.

The wording is narrow. Article 111(4) concerns providers only and paragraph 2 only. There is no transitional period for informing people that they are interacting with an AI system under paragraph 1, for informing them about emotion recognition and biometric categorisation under paragraph 3, or for deep fake disclosure by deployers under paragraph 4; those obligations have applied since 2 August 2026 regardless of when a system was placed on the market. Passing the December date on internally as “Article 50 only applies from December” misstates the law. An overview of all paragraphs is in the article on the Article 50 transparency duties.

The wording leaves open when a revised version of a legacy system counts as newly placed on the market. Under Article 3(9), placing on the market means “the first making available of an AI system or a general-purpose AI model on the Union market”. Recital 39 formulates a rule based on the type and model of a system only for the transitional provision in Article 111(2) on high-risk systems, not for paragraph 4. From 2 December 2026 the question loses relevance, since all systems must comply from then on; until then, the date of first placing on the market should be evidenced for each system.

Article 50(2) does not prescribe the technical form of the marking in detail. Under Article 50(7), as recast by Regulation 2026/1744, the Commission encourages and facilitates codes of practice at Union level for the detection, marking and labelling of artificially generated or manipulated content. If it deems a code inadequate, it may adopt an implementing act specifying common rules for the implementation of the obligations in paragraphs 2 and 4. The Code of Practice on Transparency of AI-Generated Content exists in its final version (closing plenary on 10 June 2026); the Commission and the AI Board have confirmed it as an adequate voluntary tool to demonstrate compliance (https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content).

The transitional period under Article 111(4) covers only machine-readable marking under Article 50(2), and only providers of systems placed on the market before 2 August 2026. Article 50(1), (3) and (4) have applied since 2 August 2026 without deferral. Content generated before 2 August 2026 does not have to be labelled retroactively according to the Commission’s FAQ; the Commission encourages deployers to do so where possible (see [European Commission, Transparency obligations under Article 50 AI Act, FAQ, last update 24 July 2026](https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act)).

Fines: two different tiers

The two December changes sit in different tiers of Article 99. Non-compliance with the prohibitions in Article 5, and therefore with points (ba) and (bb), is subject under Article 99(3) to administrative fines of up to EUR 35 million or, for undertakings, up to 7 % of total worldwide annual turnover for the preceding financial year, whichever is higher. Breaches of the transparency obligations in Article 50 fall under Article 99(4), point (g): up to EUR 15 million or up to 3 %.

Two rules of differing reach apply to smaller companies. For SMEs, including start-ups, Article 99(6) caps each fine under paragraphs 3, 4 and 5 at whichever amount is lower. For small mid-cap enterprises, defined in Article 3(14b) by reference to Recommendation (EU) 2025/1099, the new Article 99(6a) applies the lower amount only to fines under paragraphs 4 and 5. A small mid-cap infringing Article 5 is therefore not covered by that reduction, whereas one infringing Article 50 is.

The detailed penalty rules are laid down by the Member States under Article 99(1), taking into account the interests of SMEs and small mid-caps. Recital 15 of Regulation 2026/1744 also makes clear that the new prohibitions do not preclude prosecution under other law, including criminal law, and that Member States must respect the ne bis in idem principle where the same conduct is sanctioned twice.

InfringementProvisionMaximumSMEs (para. 6)Small mid-caps (para. 6a)
Art. 5(1), point (ba) or (bb)Art. 99(3)EUR 35 million or 7 % of turnover, whichever is higherlower amountnot covered: higher amount
Art. 50(2), including after Art. 111(4) expiresArt. 99(4), point (g)EUR 15 million or 3 % of turnover, whichever is higherlower amountlower amount

Before 2 December: six checks

The following steps follow directly from the provisions examined above. They do not replace a legal assessment of individual cases, but they structure the work for the remaining weeks.

  1. Inventory generative functions: record every system in the estate that generates or manipulates image, video, audio or text content, including embedded third-party models. For each system, the role (provider or deployer), the intended purpose and the date of first placing on the market belong in the [AI inventory](/templates/ai-inventory).
  2. Document the Article 5(1a) assessment: for every system able to generate or edit realistic depictions of people, record whether the prohibited outcome is the intended purpose, whether it can occur foreseeably and reproducibly, and which safeguards are in place. The list of measures in recital 12 works as a checklist.
  3. Make misuse correctable: Article 5(1a), point (a)(ii), requires safeguards that reliably prevent prohibited outcomes and correct observed or reported misuse. The reporting route, ownership and response process should be settled and tested before the date.
  4. Set usage rules for deployers: the internal AI use policy states that generating material within the meaning of points (ba) and (bb) is forbidden and that safeguards must not be circumvented. This links the prohibitions to the AI literacy measures under Article 4; a [skills matrix](/templates/skill-matrix) shows the training status per role.
  5. Close the marking gap for legacy systems: for every generative system placed on the market before 2 August 2026, check whether its outputs are marked in a machine-readable format and detectable as artificial. If not, 2 December 2026 is the latest date. Where an exception under Article 50(2) is relied on, the reasoning belongs on file.
  6. Clarify the supply chain: where a third-party model is embedded, the contract should state who delivers which safeguards and which marking. Whether your own organisation becomes a provider in the process is governed by Article 3(3): a provider is anyone who develops an AI system or has it developed and places it on the market or puts it into service under its own name or trademark.

Where the date sits in the AI Act calendar

2 December 2026 is an intermediate date in a chain that runs to the end of 2030. The table shows the AI Act application dates as amended by Regulation 2026/1744, each with its provision. All dates are also in the EU deadline calendar and in the regulatory cockpit, alongside NIS2, DORA, CRA and the Data Act; the EU regulation hub collects the entry points.

The list of prohibitions is not closed. Under Article 112(1), the Commission assesses once a year whether the list in Article 5 needs amending. According to the consolidated text, Regulation 2026/1744 is so far the only act amending the AI Act; recital 10 grounds the two new prohibitions in technological and societal developments since the AI Act was adopted. Which dates apply to your own systems can be assessed with the AI Act check.

DateWhat appliesProvision
2 February 2025Chapters I and II: general provisions including AI literacy (Art. 4) and prohibitions (Art. 5, excluding points (ba) and (bb))Art. 113, third para., point (a)
2 August 2025Chapter III Section 4, Chapter V (general-purpose AI models), Chapters VII and XII and Art. 78, excluding Art. 101Art. 113, third para., point (b)
27 July 2026Regulation (EU) 2026/1744 enters into force; Articles 102 to 110 applyArt. 4 Reg. (EU) 2026/1744; Art. 113, third para., point (d)
2 August 2026General date of application, including Art. 50Art. 113, second para.
2 December 2026Art. 5(1), points (ba) and (bb), Art. 5(1a) and (1b); Art. 50(2) for systems placed on the market before 2 August 2026Art. 113, third para., point (a); Art. 111(4)
2 August 2027General-purpose AI models placed on the market before 2 August 2025Art. 111(3)
2 December 2027High-risk systems under Art. 6(2) and Annex III (Chapter III Sections 1 to 3)Art. 113, third para., point (c)(i)
2 August 2028High-risk systems under Art. 6(1) and Annex IArt. 113, third para., point (c)(ii)
2 August 2030Legacy high-risk systems intended to be used by public authoritiesArt. 111(2)
31 December 2030AI components of the large-scale IT systems listed in Annex X, placed on the market or put into service before 2 August 2027Art. 111(1)

Sources and status

This article reflects the position as at 30 October 2026; all sources were accessed on that day. EUR-Lex describes the consolidated text as “meant purely as a documentation tool” with “no legal effect”; the acts published in the Official Journal are binding. This article is for orientation and does not constitute legal advice; the wording published in the Official Journal prevails.

  • Regulation (EU) 2024/1689 (Artificial Intelligence Act) of 13 June 2024, OJ L of 12.7.2024, Articles 2, 3, 4, 5, 50, 99, 111, 112 and 113: https://eur-lex.europa.eu/eli/reg/2024/1689/oj
  • Regulation (EU) 2024/1689, consolidated version of 27.7.2026 (documentation version, CELEX 02024R1689-20260727): https://eur-lex.europa.eu/eli/reg/2024/1689/2026-07-27/eng
  • Regulation (EU) 2026/1744 (Digital Omnibus on AI) of 8 July 2026, OJ L of 24.7.2026, Article 1 points 7, 38, 39 and 40, Article 4, recitals 10 to 16, 38 and 39: https://eur-lex.europa.eu/eli/reg/2026/1744/oj
  • Directive 2011/93/EU of 13 December 2011, OJ L 335 of 17.12.2011, p. 1, Article 2, points (c) and (e): https://eur-lex.europa.eu/eli/dir/2011/93/oj

Key Takeaways

  • From 2 December 2026 two new prohibitions apply: Article 5(1), first subparagraph, point (ba) (non-consensual intimate depictions of identifiable persons) and point (bb) (material within the meaning of Article 2(c) and (e) of Directive 2011/93/EU); the date derives from Article 113, third paragraph, point (a).
  • Providers infringe only if the prohibited outcome is the intended purpose, or occurs foreseeably and reproducibly without reasonable and adequate safeguards (Article 5(1a), point (a)).
  • Deployers infringe only if they use the system for the purpose of generating or manipulating such material (Article 5(1a), point (b)); according to recital 12, accidental generation is not covered.
  • For generative systems placed on the market before 2 August 2026, the transitional period for machine-readable marking under Article 50(2) ends on 2 December 2026 (Article 111(4)). Article 50(1), (3) and (4) have no transitional period.
  • Infringements of the new prohibitions fall in the highest fine tier (Article 99(3): up to EUR 35 million or 7 %); the reduction for small mid-caps under Article 99(6a) does not apply to them.

Frequently Asked Questions

Two things. First, the new prohibitions in Article 5(1), first subparagraph, points (ba) and (bb), become applicable, together with the limiting provisions in Article 5(1a) and (1b); the date follows from Article 113, third paragraph, point (a). From then on, AI systems that generate or manipulate realistic intimate depictions of identifiable persons without their explicit consent are prohibited, as are systems that generate or manipulate material within the meaning of Article 2(c) and (e) of Directive 2011/93/EU. Second, under Article 111(4), the transitional period ends for providers of generative systems placed on the market before 2 August 2026: they must have implemented machine-readable marking under Article 50(2) by 2 December 2026.

No. Under Article 5(1a), point (a), placing on the market or putting into service is prohibited only where generating such material is the intended purpose of the system, or where it is a reasonably foreseeable and reproducible outcome and the system lacks reasonable and adequate safeguards. Recital 12 states explicitly that the prohibition should not prevent the development of the technical capabilities of generative systems, and gives examples of safeguards such as data cleaning, refusal training, output controls, filters and notice and action mechanisms.

No. Article 50 has applied since 2 August 2026 (Article 113, second paragraph). The transitional period under Article 111(4) covers only machine-readable marking under Article 50(2), and only providers of systems placed on the market before 2 August 2026. Informing people that they are interacting with an AI system under paragraph 1, informing them about emotion recognition and biometric categorisation under paragraph 3, and deep fake disclosure by deployers under paragraph 4 apply without any transitional period.

As a deployer within the meaning of Article 3(4), an organisation is affected by the new prohibitions if it uses a system for the purpose of generating or manipulating material under point (ba) or (bb) (Article 5(1a), point (b)). The marking obligation under Article 50(2), by contrast, applies to providers. For deployers, the deep fake disclosure obligation under Article 50(4) has applied since 2 August 2026. In practice this means clear usage rules, a ban on circumventing safeguards and documented training under Article 4.

For infringements of Article 5, and therefore of points (ba) and (bb), Article 99(3) provides for fines of up to EUR 35 million or up to 7 % of worldwide annual turnover, whichever is higher. Infringements of Article 50 fall under Article 99(4), point (g), with up to EUR 15 million or 3 %. For SMEs, Article 99(6) applies the lower amount in each case. For small mid-cap enterprises, the reduction under Article 99(6a) covers only paragraphs 4 and 5, and so not infringements of Article 5. The detailed penalty rules are laid down by the Member States under Article 99(1).

No. EUR-Lex expressly describes the consolidated version of 27 July 2026 as a documentation tool with no legal effect. The binding texts are the acts published in the Official Journal, here Regulation (EU) 2024/1689 of 13 June 2024 and the amending Regulation (EU) 2026/1744 of 8 July 2026. For day-to-day work the consolidated version is still the practical basis, because it combines both texts; in case of doubt, the Official Journal wording should be consulted.

EU AI ActAI Act Dezember 2026Artikel 5Artikel 50Verordnung 2026/1744Regulatory Deadlines

Ready for Your Assessment?

Use our interactive templates to measure your IT organization's maturity — with automatic scores, AI-powered recommendations, and professional PDF reports.