In agent theory, epistemic trust is used to infer beliefs, for example by filtering out the information the agent receives from untrustworthy agents. Moreover, trust itself can be inferred from other information. We introduce a simple information filtering architecture that clearly distinguishes the relation between the two kinds of inference. We provide a logical analysis of the architecture, based on a new family of input/output logics. We then explore information filtering and belief manipulation within this formal framework. Our key finding is that with this architecture, some of the widely debated logical rules for trust inference are redundant with respect to information-filtering mechanisms and some others are redundant with respect to belief manipulation.

Inspired by revealed preference in economics, we study revealed epistemic trust: an agent’s (dis)trust in an information source is typically hidden, while her accept/reject behavior leaves observable traces. We model such traces by an acceptance function that maps each reported set of formulas to the subset the agent accepts. We develop two complementary models: a white-list mode, where acceptance is supported by trusted information in the report, and a black-list mode, where acceptance avoids distrusted patterns via a cautious remainder-set/full-meet construction. For both modes, we provide postulate-based representation theorems and show how canonical "revealed" trust and distrust cores can be reconstructed from the acceptance function itself.

This paper explores how to fulfill the right to explanation and support the right to challenge in automated decision-making systems by integrating symbolic methods with Large Language Models (LLMs). In cases involving automated decisions based on conflicting arguments, we first model the situation using an abstract argumentation framework. We then apply grounded semantics and discussion games to guide the explanation of the decision and support the right to challenge. Specifically, we prompt OpenAI’s flagship model (o1) to perform these reasoning steps and generate corresponding natural-language explanations. Finally, we ask the model to identify which argument would need to be modified to alter the decision, based on the formal reasoning behind the explanation. To assess the quality of the explanations, we use several state-of-the-art LLMs as evaluators. We compare three types of explanations produced by o1 with a set of criteria: those based on grounded semantics, discussion games, and a baseline explanation (in which o1 generates an explanation without any formal symbolic constraints). The results indicate that explanations based on discussion games are rated higher than those based on grounded semantics, which in turn outperform the baseline explanations. We also discuss the "right to challenge" aspect, showing that explanations based on discussion games effectively identify which arguments can be challenged to alter the decision. Overall, our findings suggest that formally guided LLMs can better fulfill the right to explanation and support the fulfillment of the right to challenge. This supports the view that integrating sub-symbolic, data-driven generative AI with symbolic, knowledge-driven AI is a fruitful way to achieve transparent AI systems that align with our societies’ legal requirements regarding digitalization.

We present a rights-first model of contrary-to-duty (CTD) reasoning with two remedial regimes and a revision track. In the CTD-Claim regime, when a primary duty is not fulfilled and no exception applies, a remedial claim detaches automatically, without any recognition act. In the CTD-Power regime, a remedial claim arises only if the rights-holder exercises a recognition power; until then there is no recognised violation and no remedial duty. Revision-of-Duty (RoD) is an alternative discretionary power that adapts the primary duty without recognising a violation, keeping the purpose aligned and avoiding sanctions. Under CTD-Power, revision competes directly with recognition on the same case. We develop a model for agreements, give concise dynamic-logic-style specifications of guards and acts, implement an institutional rights system that executes these specifications over live Hohfeldian bundles (with per-case exclusivity, exception handling and provenance), and show how agentic AI can use reasons to choose among the admissible acts within the rights-first framework.

Normative stance underlies decisions in law, legal reasoning, policy, and safety-critical settings. A model’s judgment of what is permissible vs. impermissible often determines its downstream behavior. We study how to steer a language model’s normative stances at inference time by adding a tiny, contrastive perturbation to the last-token neural activation in late MLP layers (contrastive last-token steering). For each normative prompt, we construct a contrast direction by comparing its last-token activation to that of a minimally edited variant that implies a more permissive normative stance (e.g., “acceptable” rather than “wrong”). During generation, we add this vector at the last token; a single strength parameter α controls how strongly and in which direction we push the model’s stance (permissive vs. restrictive). Impact is measured as the change in a next-token logit margin between permissive and restrictive continuations. To avoid overclaiming, we calibrate a threshold τ on neutral controls (same layers, tempered strengths with |α|≤1) and count success only when the shift exceeds τ in the expected direction. We also assess specificity by verifying that, on neutral control prompts, steered outputs exactly match unsteered baselines. Beyond component-level tests, we probe neuron-level locality by steering only the top-k contrastive neurons (ranked by last-token contrast) and confirming reversibility on our test set: +α produces the shift and -α reverses it. The method is training-free, uses standard forward hooks, and we report pilot results on Llama-3-8B-Instruct.

The advent of Distributed Ledger Technologies (DLTs) has paved the way for a new paradigm of traceability in all information systems areas. In the context of decision-making processes, however, DLTs are generally used only to trace the end results. In this work we argue that a reasoning system can be put in place for making these decisions, in order to enhance auditability, transparency, and finally to provide explainability. We propose the Intelligent Human-input-based Blockchain Oracle (IHiBO), a cross-chain oracle that enables the execution and traceability of formal argumentation and negotiation processes, involving the intervention of human experts. We take as reference the decision-making processes of fund managements, as trust is of crucial importance in such ``trust services''. The architecture and implementation of IHiBO are based on leveraging two-layer DLTs, smart contracts, argumentation and negotiation in a multi-agent setup. Finally, we provide some experimental results that support our discussion, namely that in the use-case we have considered our methodology can increase trust from principals to trusted services.

Abstract agent argumentation frameworks extend Dung’s theory with agents, and in this paper we study four types of semantics for them. First, agent defense semantics replaces Dung’s notion of defense by some kind of agent defense. Second, social agent semantics prefers arguments that belong to more agents. Third, agent reduction semantics considers the perspective of individual agents. Fourth, agent filtering semantics are inspired by a lack of knowledge. We study five existing principles and we introduce twelve new ones. In total, we provide a full analysis of fifty-two agent semantics and the seventeen principles.

The theory of formal argumentation distinguishes and unifies various notions of attack, support and preference among arguments, and principles are used to classify the semantics of various kinds of argumentation frameworks. In this paper, we consider the case in which we know that an argument is supporting another one, but we do not know yet which kind of support it is. Most common in the literature is to classify support as deductive, necessary, or evidentiary. Alternatively, support is characterized using principles. We discuss the interpretation of support using a legal divorce action. Technical results and proofs can be found in an accompanying technical report.