Publication

Research on Open Source LLM Safety at HICSS 2026

From January 6-9, 2026, TWON researcher Simon Münker presented his paper at the Hawaii International Conference on System Sciences (HICSS), one of the leading international conferences in the field of information systems and digital innovation.

The paper addresses societal risks associated with open source Large Language Models and evaluates the effectiveness of existing safety and guardrail mechanisms. Together with his co author Fabio Sartori, Simon Münker received the Best Paper Award for this research.

The study systematically examines guardrail vulnerabilities across seven widely used open source LLMs. Using advanced natural language processing classification methods, it identifies recurring patterns of harmful content generation under adversarial prompting. These vulnerabilities were first observed during earlier research activities within the TWON project, where initial experiments revealed persistent weaknesses in model safety mechanisms.

The findings show that several prominent models consistently produce content classified as hateful or offensive. This raises concerns about the potential implications of open source LLMs for democratic discourse and social cohesion. In particular, the results challenge public safety assurances by model developers and point to discrepancies between stated safeguards and observed model behavior.

The research contributes to ongoing discussions on responsible AI development and the governance of AI systems that shape online communication and public discourse. It underlines the need for more robust, transparent and empirically tested safety mechanisms in open source AI ecosystems.

The paper was presented as part of the Digital Democracy Minitrack at HICSS 2026.

“Navigating the Digital Age” – New Book Co-Edited by TWON-Researchers

TWON-researchers Damian Trilling of the University of Amsterdam and  Ljubiša Bojić of the University of Belgrad have co-edited the book: “Navigating the Digital Age: An In-Depth Exploration into the Intersection of Modern Technologies and Societal Transformation” alongside Jörg Matthes and Simona Žikić.

The book approaches the complex issues at the intersection of new technology and society in four sections focusing on communication technology, the digital transformation of society, the relationship of technology, democracy and security, and the impacts of generative artificial intelligence (AI) in the arts respectively.

A range of international scholars and academics contributed to this book, covering diverse issues arising from the rapid dissemination of new technologies and changing digital landscapes. Ljubiša Bojić and Damian Trilling together with their co-editors contributed to the books preface. Additionally, Ljubiša Bojić co-authored a text on AI in governement decision making processes.

Collectively, these works enhance our understanding of the intricate relationship between technology and society, offering a critical analysis of the ethical, moral, and cultural impacts of rapid digital change.

The book builds on the insights gathered in projects such as TWON that set out to investigate the impacts of new technologies on our societies, leveraging new technologies to better understand the digital transformation we are living through and helping to build more democratic digital environments.

The book has been published by the Institute for Philosophy and Social Theory at the University of Belgrad.

find out more about the book
Downolad the book

On the Ethics of Using TWONs: TWON Policy Brief #1

Researchers of the TWON consortium have achieved another milestone regarding the communication of results and insights of our EU-funded research project. As we are breaking new ground at the intersection of diverse academic discplines, we are confronted with new ethical questions relating to our work. A Digital Twin of an Online Social Network (TWON) may be a powerful instrument for researching and regulating digital public spaces. Yet, the vast and detailed datasets on user behaviour also pose significant ethical risks. TWONs could be used to undermine user interests and democratic norms. Important questions arise with regard to the technological tools we are developing: Who should have access to these tools? How should these tools be regulated?

As scholars researching the relationship of social and political dynamics on one hand and digital technologies and communicative settings on the other hand, we do not want to shy away from these important normative questions: The answers to these questions may well determine the societal impact of our joint research project. Therefore, we aim to address these crucial normative and ethical questions heads on. Our first TWON Policy Brief, thus, focuses on these questions, emphasizing the importance of ethical considerations in our research and the dissemination of our findings. The policy brief sketches out possible approaches to TWONs, different governance and regulation frameworks as well as their risks and benefits.

The full report can be found here.

2024-10-28_TWON-policy-briefDownload

 

A Call to Action: New Paper on Youth Online Mental Health

The World Health Organization (WHO) has put forward guiding principles on online mental health for young people. In a newly published study, TWON-researchers integrate the WHO framework into a European context. They emphasize the impact of digital platforms and online social networks (OSNs) on mental health. The COVID-19 pandemic has amplified both the usage of digital technologies and a mental health crisis that is affecting young people in particular. Traditional services are increasingly losing touch with the young target group. Substantial challenges arise from the need to provide accessible and flexible mental health support while ensuring adherence to evidence-based guidelines.

The authors stress the special character of the European context: Typically, most young Europeans are growing up as digital natives, given the broad accessibility of advanced technology. Yet, there are important regional differences concerning the approach to mental health and the influence of social stigma related to the concept. Taking into account socioeconomic discrepancies and varying degrees of media literacy across Europe, this stresses the need for culturally sensitive mental health support. So far, online mental health services are often provided by independent organizations. But like other health-related services and products, the sector is in need of a sound regulatory framework to ensure both the quality and accessibility of services.

This highlights the need to gain a better understanding of platform mechanisms and the effects of algorithmic hyper-personalization and, therefore, the importance of our joint project TWON. Based on their research, the authors developed five core recommendations:

1. The specific European context must be considered.

2. Quality and accessibility of services must be balanced carefully.

3. Involving young people and families, leveraging their knowledge while enhancing digital literacy, is key.

4. The rise of AI must be addressed, concerning both the potential of AI for developing mental health support and the risks of biases and misinformation.

5. A regulatory framework must be developed to ensure safety, effectiveness, and ethical standards.

If you want to learn more about the issue, you can access the full article here!

New Publication on the Theoretical Foundations of Building a TWON

In a study recently published in the European Journal of Futures Research, Ljubiša Bojić (JSI) and his team explored a multidisciplinary approach to testing and aligning artificial intelligence (AI), with a special focus on large language models (LLMs). They investigated a simulation-based multi-agent system replicating a real-world environment. In this case, the paper focuses on a digital replication of a city, populated by “digital citizens” that are simulating complex social structures and behaviours. This theoretical work is vital for our joint research project and provides valuable insights for the construction of a Twin of an Online Social Network (TWON).

The article identifies theoretical perspectives from diverse disciplines that can contribute to developing AI that is socially responsible and aligned with human values. The strength of the article lies in its ability to bring together theoretical perspectives from fields as diverse as sociology, social psychology, computer science, physics, biology, and economics. In this multidisciplinary approach, the study resembles the TWON project as a whole – bringing together researchers from across Europe and from diverse academic backgrounds.

If you want to learn more, you can access the full article here.

Refining Deliberative Standards for Online Political Communication: Introducing a Summative Approach to Designing Deliberative Recommender Systems

How can Social Media best foster democratic debates? In a new paper, TWON’s very own Sjoerd Stolwijk, Michael Heseltine, Corinna Oschatz, and Damian Trilling challenge the notion that a perfect debate on social media platforms is a desirable outcome.

Over the last few years, diagnoses that the democracies of post-industrial nations of the global north are in decline have spread like proverbial wildfire. Often times, the way that social media continues shaping our lives and interactions with one another is held at least partially accountable for its ostensible effect on larger societal structures.

Contrasting the established “additive” paradigm of deliberative democracy research, Stolwijk, Heseltine, Oschatz and Trilling “propose an alternative conception of debate quality for online platforms, based on the recently proposed systematic, summative approach to deliberative democracy”. Using this approach, the researchers shift the focus from singular factors of deliberation to more systemic interdependences of deliberative factors and their effects on society at large.

Starting with an outline of critiques of the additive approach and its unitary implications, Stolwijk et al. inquire into the effects of deliberation indicators on one another as opposed to their effects on a monolithic deliberation. Continuing, they pose the question of exactly where online platforms fit into the larger system of deliberative democracy and whether they can be placed within a ‘micro-meso-macro’-layer approach.

On the base of this contextualisation, the authors move on to propose the ‘summative approach’ to deliberative democracy and online platforms’ role within it, arriving at the proposition that online communication should be viewed as complementary to existing forms of deliberation rather than a replacement.

“when facilitating debate between citizens, instead of aiming for civil conversation, it might be better for (macro) deliberative democracy, if in some cases people are allowed some incivility to make suppressed voices heard or to create a communicative environment where some might feel more at home, where they feel they don’t need to be eloquent and highly educated to be allowed to speak up”

The authors included a comparative table containing both ‘additive’ and ‘summative’ indicators for successful deliberation, both as an overview and an invitation for other researchers to contribute and improve. Concluding remarks reiterate the paper’s findings and summarise the scientific value of the ‘summative’ approach.

In an effort to promote inclusivity and accessibility, papers published by the TWON consortium are published as open-source. So check out our paper for free, published here.