Dev.to AI 🤖 Ai 👁 0 📖 5 min read

New AI Act Prohibitions Take Effect December 2

Originally published at https://charz.ai/blog/new-ai-act-prohibitions-december-2026 by Char-Z AI. What Are the New AI Act Prohibitions? Two new prohibited practices take effect on December 2, 2026, added to Article 5 o

Originally published at https://charz.ai/blog/new-ai-act-prohibitions-december-2026 by Char-Z AI.

What Are the New AI Act Prohibitions?

Two new prohibited practices take effect on December 2, 2026, added to Article 5 of the EU AI Act. They join the original February 2, 2025 prohibited practices and — unlike the high-risk regime — they are not deferred by the Omnibus's timeline changes.

The short version: Article 5 lists practices that are forbidden outright. No high-risk registration, no conformity assessment, just a ban. The original set covered manipulative techniques, exploitation of vulnerabilities, social scoring, certain biometric categorisations, emotion inference in the workplace and education settings, and real-time remote biometric identification in public spaces. The two new limbs target non-consensual synthetic sexual media and child sexual-abuse material. Both apply from December 2, 2026 (Regulation (EU) 2024/1689, Art. 5(1)(ba) and (bb), as amended).

The First Limb: Non-Consensual Intimate and Sexually Explicit Material

Article 5(1)(ba) prohibits the placing on the market, the putting into service for first use directly to a deployer, own use in the Union for its intended purpose, 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
  • an identifiable natural person engaged in sexually explicit activities ,

without that person's freely-given, specific, informed, unambiguous and explicit consent (Regulation (EU) 2024/1689, Art. 5(1)(ba), as amended).

The operative test is consent. It is not detectability, not provenance, and not whether the output looks artificial. What the provision asks is whether the person depicted consented to that generation or manipulation — and the standard is deliberately demanding, requiring that consent to be freely given, specific, informed, unambiguous and explicit. A system capable of producing realistic synthetic imagery of a real person sits inside the prohibition unless it holds that person's consent on a record it can produce.

Note the modality. This limb is framed in terms of realistic images, videos, audio or similar material — synthetic media, not text output.

The Second Limb: Child Sexual-Abuse Material and Performance

Article 5(1)(bb) prohibits the same three acts — placing on the market, putting into service for first use, own use in the Union, or use — where the AI system generates or manipulates material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU (Regulation (EU) 2024/1689, Art. 5(1)(bb), as amended).

Those points of the EU's child-protection directive cover child sexual-abuse material and performance involving a child. The limb carries one qualifier: it does not bite where a "without right" defence applies under national law, so existing domestic-law defences for possessing or accessing such material continue to operate.

Who It Applies To

  • Providers placing an AI system on the EU market — Art. 5 prohibitions apply from Dec 2, 2026.
  • Providers putting a system into service for first use directly to a deployer in the Union — bound from Dec 2, 2026.
  • Any actor using such a system for own use in the Union for its intended purpose — bound from Dec 2, 2026.
  • Deployers using such a system in the EU — subject to the operational duties where the use falls within the prohibited scenarios.

How This Relates to the Watermarking Deadline

The two land on the same day, but they are separate obligations and discharging one does not discharge the other.

Article 50(2) requires providers of certain AI systems to mark synthetic outputs in a machine-readable format so they are identifiable as artificially generated or manipulated. Article 5(1)(ba) and (bb) prohibit specified categories of content outright. Marking an output does not make prohibited content lawful — the marking duty is a transparency measure, while the prohibition turns on the consent of the person depicted and on child-protection law. A system that satisfies every Article 50(2) requirement can still be prohibited under Article 5.

The reverse also holds: suppressing a prohibited use does not discharge the Article 50(2) marking duty for the outputs you remain permitted to produce. Treat them as two separate lines of work with two separate evidence trails.

What Organizations Should Do

  • Treat December 2, 2026 as a live date. The prohibitions are enforceable on the same day the marking grace period ends — neither waits for the December 2027 high-risk regime.
  • Establish a consent basis for any synthetic depiction of a person. If your systems can generate realistic images, video or audio of identifiable individuals' intimate parts or sexually explicit activity, you need a demonstrable, specific and explicit consent record per subject, plus a working route to withdraw it. Treat the consent store as the compliance artifact — not the model configuration.
  • Audit for child sexual-abuse material generation paths. Check your stack for generation, transformation and search features that could surface material within Article 2(c) or 2(e) of Directive 2011/93/EU, and confirm your moderation and reporting path handles the "without right" defence correctly rather than over-blocking lawful material.
  • Separate marking work from prohibition work. Do not treat Article 50(2) marking as your Article 5 mitigation. They have different legal tests, different scopes and different evidence.
  • Re-read your acceptable-use terms. The prohibition is provider- and deployer-facing. Your acceptable-use policy, vendor documentation and model cards should state that non-consensual intimate and sexually explicit material, and child sexual-abuse material, are out of scope for the service.
  • Don't confuse the deadline sets. These are 2026 obligations; the Annex III high-risk regime that moved to December 2, 2027 is a separate, future compliance project.

Frequently Asked Questions

When do the new Article 5 prohibitions take effect?

December 2, 2026. Both limbs were added by amendment and apply from that date under Article 113(a) (Regulation (EU) 2024/1689, Art. 5(1)(ba) and (bb), as amended).

What is now banned under Article 5?

Two things. First, AI systems that generate or manipulate 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. Second, AI systems that generate or manipulate material or performance within Article 2, points (c) and (e) of Directive 2011/93/EU — child sexual-abuse material and performance — except where a "without right" defence applies under national law. They join the original prohibited practices that have applied since February 2, 2025.

Does watermarking make content compliant with the new prohibition?

No. Article 5(1)(ba) turns on whether the person depicted consented, and Article 5(1)(bb) turns on child-protection law. Neither test looks at whether the output is detectable or machine-readably marked. Article 50(2) marking is an independent transparency duty; satisfying it does not make a prohibited use lawful, and suppressing a prohibited use does not satisfy it.

Does the first limb cover generated text?

Not on its face. Article 5(1)(ba) is framed in terms of realistic images, videos, audio or similar material — synthetic media rather than text output. Text generated by an AI system is addressed elsewhere, under Article 50, where an AI system generates public-interest text; that is a disclosure duty, not this prohibition.

What are the penalties for a prohibited AI practice?

Prohibited practices sit in the top penalty tier: up to 35 million EUR or 7% of global annual turnover (European Commission, 2024).

Sources

European Commission. (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council. *Official Journal of the European Union*. https://eur-lex.europa.eu/eli/reg/2024/1689
European Commission. (2026). Regulation (EU) 2026/1744 — Digital Omnibus on AI. *Official Journal of the European Union*.
📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.