No Certificate, No Categorical Speech Act
A Brouwerian Assertibility Constraint for Public Machine Speech
Generative AI can turn uncertainty into verdicts that sound authoritative. The paper proposes that in high-stakes domains, AI systems may assert or deny a claim only with a public, contestable certificate, and otherwise answer “Undetermined”.
arXivFrom Awareness to Action?
The Impact of CO2 Emission Feedback on Student LLM Usage
Students at LMU Munich saw the estimated CO₂ emissions of their LLM use while they worked. First findings suggest that this real-time feedback can reduce emissions and tokens per prompt.
ACM CHI EA ’26The Continuum of Law
An Intuitionistic Social Ontology of Normative Systems
Law is understood as an open-ended normative system whose identity is constituted through its history of construction. Drawing on Brouwer’s intuitionism, the paper distinguishes historically realized legal pathways from their changing operative force. The resulting ontology offers a framework for legal AI that preserves legal history while representing repeal, suspension, and the self-revision of legal order.
PhilArchiveNo-Miracles Argument of Law
Jurisprudence Meets Structural Realism
Law is read as a structured field in which meaning arises from relations such as priority, exception, competence and authority. Inspired by structural realism, the paper develops structural normativism, a foundation for legal AI that makes these relations explicit.
SpringerFrom Prompt to Propaganda
Generative AI and the New Disinformation Lifecycle. Workshop Report, Ecosystem Cartography and Research Agenda
The report documents an international workshop at the INFORMATIK FESTIVAL 2026 in Dresden on how generative AI changes information manipulation. It maps research and response capabilities in Germany, France and Ukraine and sets out thirteen research challenges.
ZenodoWhat Is Said and How It Is Hidden
Measuring Political Suppression in Large Language Models
This ongoing research project investigates subtle forms of information suppression in large language models, with a scientific publication in preparation. AI systems may provide fluent, seemingly informative answers while omitting crucial facts or subtly distorting information. The research develops a multidimensional framework to identify and measure such patterns by examining the relationship between topic sensitivity, linguistic elaboration, and factual alignment. The goal is to make hidden forms of information suppression more transparent and measurable, contributing to more accountable and trustworthy AI systems.
2026