Are Softfakes Really “Soft”? Semantic, Ethical, and Legal Perspectives on Deepfakes and AI-generated Visualizations in Political Campaigning

Abstract This paper critically assesses a new trend in political communication: The rise of deepfakes and AI-generated visualizations to promote candidates or parties in political campaigns – and the term “softfakes” coined to describe this phenomenon. We systematize current use cases of generative audio-visual AI for campaigning, namely multilingual outreach, testimonials of support (including by deceased people), likeable depictions and flawless avatars of candidates, AI-mediated two-way communication between politicians and citizens, (hyper-)personalized political advertising, campaigning under (alleged) repression, and illustrating political campaigns. We argue that deeming such uses of deepfakes and less sophisticated generative AI content “soft” is unjustified from a semantic, ethical, and democracy-theoretical perspective since they bear great potential for manipulation and deception. Moreover, they contribute to fragmenting the political debate and raise serious ethical concerns related to human dignity. Besides, we argue that accepting the term “softfake” could contribute to public and political indifference to digital manipulation, thereby obstructing regulation. Since the use of generative AI for audio-visual political campaigning will arguably become the “new normal”, political actors, society, and researchers must consider e.g., adapting regulatory frameworks for political advertising, establishing industrial or political codes of conduct, and enhancing public media literacy and resilience to digital manipulation.