Introduction

AI systems are increasingly deployed in contexts where their behaviour should be constrained not only by technical requirements but also by laws, regulations, standards, ethical principles, organisational policies, and social norms. Ensuring that such systems behave in accordance with these norms raises fundamental challenges across the entire AI lifecycle. The workshop addresses three interconnected questions: (1) How can norms be extracted, formalised, and represented faithfully? (2) How can they be incorporated into AI systems’ reasoning and decision-making? (3) How can compliance with them be verified, audited, and certified?

Recent advances in NLP and LLMs create new opportunities for automatically transforming normative texts into formal representations. However, a syntactically valid representation may omit relevant conditions, misrepresent exceptions, or introduce unintended interpretations. Establishing the faithfulness of formalised norms is therefore an essential foundation for assessing the compliance of systems that use them. Integrating these representations into AI systems introduces further challenges concerning normative reasoning, conflicts between requirements, system behaviour, and human oversight.

VLE-LEGATO will address these complementary challenges through two tracks:
* VLE (Verifying Law and Ethics) Track: Verification of Legal and Ethical Compliance. This track focuses on incorporating norms into AI architectures, reasoning, and decision-making, and on verifying, auditing, and certifying system compliance. It also addresses conflicts between legal and ethical requirements, compliance monitoring, and governance.
* LEGATO (LEGal AuTOformalization)Track: Autoformalization and Validation/Verification of Norms. This track focuses on extracting norms from natural-language and other sources, generating formal representations, and assessing whether those representations faithfully capture the relevant normative meaning and inferential behaviour. It also addresses benchmarking, debugging, human feedback, and the maintenance of normative knowledge, with particular attention to legal autoformalization.

Paper Submission:
Authors should access to a track page to which they would like to submit their papers and follow instructions provided there:

• VLE (Verifying Law and Ethics) Track: Track Page


• LEGATO (Legal Autoformalization) Track: Track Page

Topics of Interest

VLE Track: Verification of Legal and Ethical Compliance

• Normative reasoning, including obligations, permissions, prohibitions, exceptions, and conflicts
• Integration of legal and ethical norms into AI decision-making, world models, and autonomous systems
• Symbolic, machine-learning, LLM-based, and hybrid architectures incorporating normative constraints
• Specification and verification of ethical properties, including fairness, transparency, explainability, and non-discrimination
• Representation and management of conflicts between legal requirements and ethical considerations
• Formal verification, testing, counterexample generation, and debugging of AI systems subject to normative requirements
• Algorithmic auditing and compliance monitoring
• Certification, standardisation, and labelling of intelligent systems
• Practical applications, governance, regulatory frameworks, and case studies

LEGATO Track: Autoformalization and Validation of Norms

• Automatic and semi-automatic extraction, formalization, and representation of norms from statutes, regulations, contracts, judgments, policies, and other sources
• LLM-based, neuro-symbolic, rule-based, and ontology-based approaches to normative representation
• Validation of source faithfulness, including conditions, exceptions, cross-references, and temporal scope
• Semantic and behavioural equivalence of alternative normative representations
• Testing, counterexample generation, and debugging of generated formal representations
• Ambiguity, uncertainty, implicit assumptions, and alternative normative interpretations
• Human-in-the-loop workflows, interfaces, and expert feedback
• Datasets, shared tasks, benchmarks, evaluation metrics, and reproducibility
• Traceability, versioning, maintenance, and governance of normative knowledge

Joint Topics

• End-to-end methods connecting norm extraction and autoformalization with system compliance verification
• Propagation of uncertainty and interpretive choices from normative sources to system behaviour
• Traceability from source norms through formal representations to compliance evidence
• Shared benchmarks evaluating both representation faithfulness and system compliance
• Human oversight throughout the normative AI lifecycle