AI Act transparency rules face scrutiny as Commission draft full of exceptions
EU AI Act transparency rules face scrutiny as a Commission draft on deepfake disclosures includes numerous exceptions, alarming civil society and regulators.
The European Commission’s draft guideline on AI Act transparency rules aims to curb fraud from AI-generated and manipulated media by clarifying disclosure obligations for deepfakes and synthetic content. The draft, however, contains a large number of exceptions and illustrative examples that could limit when and how disclosures are required. Lawmakers, regulators and civil society groups are raising questions about whether the guidance will effectively close the transparency gaps it targets.
Commission draft and its stated purpose
The Commission circulated a draft guideline intended to explain which actors must disclose the use of AI when producing or distributing synthetic media. The guidance seeks to operationalize provisions of the AI Act related to transparency, particularly disclosures connected to deepfakes and other manipulated audiovisual material.
Officials describe the aim as protecting the public from deception while preserving legitimate uses of synthetic content such as satire, art and research. The current draft, however, illustrates numerous borderline cases and carve-outs that critics say could undermine a clear, enforceable standard.
Deepfake definition under debate
The draft does not adopt a single, rigid definition of “deepfake,” instead offering a spectrum of examples to illustrate when content qualifies as AI-generated or materially manipulated. That approach is intended to give flexibility across technologies, but it also leaves room for divergent interpretations by platforms and national authorities.
Civil society organisations warn that relying on examples rather than a compact legal definition could produce inconsistent application across member states. Experts caution that enforcement will hinge on how regulators interpret those examples in practice and whether penalties for non-compliance are applied uniformly.
Scope of disclosure obligations
The guideline attempts to delineate who must disclose AI involvement, spanning content creators, platform operators and providers of synthetic media tools. It also distinguishes between disclosures required at the point of creation and those required at the point of distribution, an important nuance for publishers and social networks.
Despite these clarifications, the draft leaves open questions about responsibility when multiple parties are involved in producing or amplifying manipulated content. Industry actors are seeking clearer rules on obligations for intermediaries, automated moderation systems and downstream publishers that re-share flagged media.
Exceptions that could narrow coverage
A central criticism of the draft is the breadth of exceptions and illustrative exemptions it proposes. Examples include scenarios flagged as parody, historical reenactment, or minor, non-deceptive edits — categories that could be interpreted widely by those seeking to avoid disclosure duties.
Regulatory lawyers say these carve-outs may reduce the practical scope of the transparency regime, especially where intent is hard to prove. Campaign groups argue that broad exceptions risk leaving the public exposed to sophisticated forgeries that are presented as authentic, while platforms gain legal wiggle room to avoid labeling problematic content.
Industry and civil society reactions
Technology companies have generally welcomed the draft’s flexibility, saying that prescriptive rules could stifle innovation and legitimate uses of generative AI. Platform operators emphasise implementation challenges, including technical detection limits and the burden of labeling legacy content.
By contrast, advocacy groups and several journalists’ organisations urge tighter language and narrower exceptions to prevent misuse. They call for robust enforcement mechanisms and clear guidance for cross-border cooperation among national authorities tasked with implementing the AI Act.
Regulatory next steps and implementation timeline
The Commission will likely revise the guideline following consultations with member states, industry stakeholders and civil society, before issuing a final text to accompany the AI Act’s enforcement framework. National regulators will then need to translate the guidance into operational enforcement practices, a process that could reveal gaps between theory and application.
Experts expect disputes over interpretation to surface early, which may prompt follow-up clarifications or litigation that will test the reach of disclosure obligations. Observers say prompt, transparent dialogue between regulators and affected sectors will be critical to achieving consistent application across the EU.
The final shape of the AI Act transparency rules will determine how effectively the EU addresses the rise of AI-enabled fraud in media, balancing protection against deception with legitimate uses of synthetic content. Stakeholders on all sides now await the Commission’s revisions and the ensuing debate in member states that will determine how those rules function in practice.
