Regulating Deep Fakes in the Artificial Intelligence Act

Abstract

The Artificial Intelligence Act (AI Act) may be a milestone in the regulation of artificial intelligence by the European Union. The regulatory framework proposed by the European Commission has the potential to serve as a global benchmark and strengthen the position of the EU as one of the main players on the technology market. One of the components of the draft regulation are the provisions on deep fakes, which include a relevant definition, risk category classification and transparency obligations. Deep fakes rightly arouse controversy and are a complex phenomenon. When leveraged for negative purposes, they significantly increase the risk of political manipulation, and at the same time contribute to disinformation, undermining trust in information and the media. The AI Act may strengthen the protection of citizens against some of the negative consequences of misusing deep fakes, although the impact of the regulatory framework in its current form will be limited due to the specificity of their creation and dissemination. The effectiveness of the provisions will depend not only on enforcement capabilities, but also on the precision of phrasing provisions to prevent misinterpretation and deliberate abuse of exceptions. At the same time, the AI Act will not cover a significant portion of deep fakes, which, due to the malicious intentions of their creators, will not be subject to the transparency obligations. This study analyses provisions related to deep fakes in the AI Act and proposes improvements that will take into account the specificity of this phenomenon to a greater extent.