bidt project on Authoritarian AI: How large language models are adapted to Russia's propaganda

In a project funded by the bidt, researchers from the Universities of Passau and Bamberg are investigating how Russia is developing its own generative AI models under strict supervision and how authoritarian data affects AI systems in democratic systems.

The Pope in a Gucci jacket, Taylor Swift as a supposed Trump supporter, a fake video by German politician Friedrich Merz: generative artificial intelligence systems can now produce deceptively real content. They are likely to fundamentally change the way information circulates in modern societies. Authoritarian regimes have long recognized the power of AI-generated images and content.

In a new project, an interdisciplinary team of researchers from the Universities of Passau and Bamberg is investigating how Russia and other autocracies regulate and control the development of large language models (LLMs). The Bavarian Research Institute for Digital Transformation (bidt) of the Bavarian Academy of Sciences and Humanities is funding the AI-PROP project on authoritarian AI and propaganda with 930,000 euros over a period of three years. It will start on April 1, 2025.

“In autocracies such as China and Russia, generative AI is now often developed under the auspices of state-controlled companies,” says project leader Professor Florian Töpfl, Chair of Political Communication with a focus on Eastern Europe at the University of Passau. “Through such measures of regulation and censorship, autocratic governments ensure that the output of AI models is in line with the political narratives and ideologies of the ruling elites.” In addition, autocrats are increasingly censoring access to the language models developed in democracies. China and Russia, for example, have banned the use of ChatGPT, the chatbot developed by US software company OpenAI.

In addition to Professor Töpfl, the computer scientist Professor Florian Lemmerich from the University of Passau and the political scientist Professor Andreas Jungherr from the University of Bamberg are also involved in the project. Against the backdrop of current developments, the researchers in the bidt project are addressing two overarching questions:

• How and with what consequences are authoritarian AI systems embedded in today's Russian society?
• What are the consequences of “authoritarian data”, i.e. data that has been manipulated by authoritarian influence and censorship, increasingly being fed into Western AI systems?

As part of four interconnected work packages, the team is investigating how systems based on language models are regulated in Russia compared to Western democracies and how the content generated by these systems differs in democratic and authoritarian states. In addition, the researchers are exploring the consequences of the rise of generative AI on public discourse in democratic and authoritarian societies.

“Ultimately, our aim with the 'Authoritarian AI' project is to strengthen the resilience of democracies against authoritarian influence by means of artificial intelligence and to make transparent how authoritarian elites manipulate AI systems,” explains Professor Töpfl. The project also includes a section on public relations and knowledge transfer, in which a guide is to be developed to build up expertise against propaganda.

About the research team

Project leader Professor Florian Töpfl holds the Chair of Political Communication with a focus on Eastern Europe and the post-Soviet region at the University of Passau. He heads the ERC Consolidator project RUSINFORM on the effects of digitalization on Russia's informational influence abroad. Professor Töpfl's research interests include the political communication of anti-democratic counter-publics in Western democracies.

Professor Florian Lemmerich holds the Chair of Applied Machine Learning at the University of Passau. His research interests include the development of machine learning methods and the quality of large language models.

Professor Andreas Jungherr holds the Chair of Political Science, in particular Digital Transformation, at the University of Bamberg. Among other things, he investigates the use of digital media by the population, political elites, media and organizations.

This text was machine-translated from German.

Symbolic picture: Adobe Stock

Principal Investigator(s) at the University	Prof. Dr. Florian Töpfl (Lehrstuhl für Politische Kommunikation mit Schwerpunkt auf Osteuropa und die postsowjetische Region)
Project period	01.01.2025 - 31.12.2027
Source of funding	BAdW - Bayerische Akademie der Wissenschaften > BAdW - Bayerische Akademie der Wissenschaften - bidt - Bayerisches Forschungsinstitut für Digitale Transformation
Themenfelder	Kommunikationswissenschaft/Publizistik, Politikwissenschaften, Politikwissenschaften allgemein, Angewandte Informatik, Publizistik und Kommunikationswissenschaft, Politikwissenschaft