EU AI Act - Synthetic Media Transparency and New Prohibitions (2 Dec 2026)

Article 50 transparency duties for providers of systems generating synthetic audio, image, video or text, plus the Digital Omnibus prohibitions on AI that generates non-consensual intimate imagery or child sexual abuse material.

JurisdictionEuropean Union
CategoryAI Regulations
StatusUpcoming
Effective date
Latest development

Analysis

The EU AI Act’s Article 50 transparency rules and the Digital Omnibus prohibitions on non‑consensual intimate imagery and child sexual abuse material (CSAM) impose mandatory content marking, user disclosure, and outright bans on certain AI practices, with key dates of 2 August 2026 and 2 December 2026 for phased application and grace periods.Article 50 official textEU Commission Transparency GuidelinesEP Omnibus Press ReleaseDigital Omnibus overview

Key Requirements

1. Machine‑readable marking of synthetic content (Article 50(2))

2. User disclosure when interacting with AI systems (Article 50(1))

3. Transparency for deepfakes and AI‑generated public‑interest publications (deployers)

4. New prohibitions on non‑consensual intimate imagery and CSAM (Digital Omnibus – Article 5 additions)

  • The Digital Omnibus on AI adds new prohibited AI practices to Article 5 of the EU AI Act targeting:
  • AI systems that generate or manipulate realistic images, video, audio or similar material of an identifiable natural person’s intimate parts or sexually explicit activities without that person’s freely‑given, specific, informed, unambiguous and explicit consent.
  • AI systems used to generate child sexual abuse material. Digital Omnibus explorer textEP Omnibus Press ReleaseCommission report on prohibitions
  • These prohibitions apply to placing such AI systems on the market, putting them into service, or using them for those purposes, with limited exceptions for lawful detection, investigation, or moderation of CSAM and consent‑based intimate content. Digital Omnibus explorer textEU Omnibus legal briefing789329_EN.pdf)

Compliance Challenges

1. Technical implementation of machine‑readable watermarking at scale

  • A widely cited difficulty is retrofitting watermarking to legacy systems already deployed before 2 August 2026, which prompted the Omnibus transitional period to 2 December 2026 for pre‑existing systems. EP Omnibus Press ReleaseDigital Omnibus adoption note789329_EN.pdf)

2. Boundary between “deepfake” and legitimate creative or satirical content

3. Detection and mitigation of non‑consensual intimate imagery and CSAM capabilities

4. Cross‑functional governance and documentation

Implementation Best Practices

1. Central AI inventory and risk classification

2. Technical implementation of watermarking and labelling

3. Safeguards against non‑consensual intimate imagery and CSAM

4. Policies, documentation, and training

  • Train staff (product, engineering, content teams) on recognizing deepfakes, prohibited content, and proper labelling practices. [EU Commission

Recent developments

  • — Article 50 transparency rules of the EU AI Act entered into force on 2 August 2026, requiring anyone creating or deploying AI-generated content in the EU to meet new marking and labelling obligations for synthetic media and deepfakes, with exemptions for purely personal, research and certain artistic uses[38]. (source)
  • — A detailed client memo explains that on 20 July 2026 the Commission published final guidelines and confirmed the Code of Practice for Article 50, clarifying that transparency duties broadly take effect from 2 August 2026, with a deferral to 2 December 2026 for marking and detection duties on generative systems already on the market and to 2 February 2027 for watermark-detection interoperability[40]. (source)
  • — The European Commission and national authorities announced the start of enforcement of the AI Act from 2 August 2026, including obligations for chatbots to disclose they are AI, mandatory labelling of deepfakes, and machine-readable marks on AI-generated or altered content to enable detection[3][21]. (source)
  • — A law firm update reports that the Commission and AI Board have confirmed the Code of Practice on Transparency of AI-Generated Content as an adequate compliance tool for Article 50, summarising two core duties: providers must machine-readably mark generative outputs and deployers must disclose deepfakes and certain public-interest AI text, enforceable from 2 August 2026[30]. (source)
  • — A feature article discusses how the EU’s deepfake labelling rules under the AI Act will require providers of major generative AI systems to ensure synthetic media is clearly recognisable as artificial, exploring technical and practical challenges and questioning whether watermarking and labelling regimes will be effective in practice[6][27]. (source)
  • — Coverage of the Commission’s July 20, 2026 guidance notes that the Guidelines and the Code of Practice together fix the technical details for Article 50 obligations, including machine-readable marking of synthetic audio, image, video and text and mandatory disclosure when deployers publish deepfakes or AI-generated text on matters of public interest, with staged deadlines up to 2 December 2026[39]. (source)
  • — Analysis of the AI Omnibus regulation describes how it amends the AI Act’s timelines by granting generative AI systems placed on the market before 2 August 2026 until 2 December 2026 to comply with Article 50(2) marking obligations and adds new prohibitions on AI systems that generate non‑consensual sexual deepfakes and child sexual abuse material[14]. (source)
  • — A June 2026 commentary details the second draft Code of Practice on marking and labelling AI-generated content, noting that Article 50 transparency obligations apply from 2 August 2026 but that the AI Omnibus introduces a grace period until 2 December 2026 for systems placed on the market or put into service before that date[37]. (source)
  • — A briefing on “significant changes” to the EU AI Act explains that lawmakers amended the schedule for mandatory marking of AI-generated content and expanded the list of prohibited AI to cover systems used to generate non‑consensual sexualised deepfakes and child sexual abuse material, reflecting growing concern over abusive synthetic media[23]. (source)
  • — Reporting on a provisional political deal in May 2026 notes that EU countries and lawmakers agreed that mandatory watermarking of AI-generated output would apply from 2 December 2026, and that the compromise responded to industry backlash while still tightening requirements around synthetic media transparency[28][16]. (source)

Related regulations

Put it into practice

Browse all regulations · Compliance deadlines · Latest updates