AI with evidence, structure and control.
We use AI where probabilistic reasoning is useful, and formal systems where determinism, validation and auditability are required. This page describes the layers that make that possible and the standards they build on.
AI with evidence, structure and control.
We use AI where probabilistic reasoning is useful, and formal systems where determinism, validation and auditability are required.
- AIextraction, classification, interpretation, assistance
- Structured regulatory modelsvalidation, consistency, provenance
- Formal rulesdeterministic execution
- Evidencesource-level traceability
AI extracts and assists. Structured models validate. Formal rules execute.
LLM-only systems
Good at interpreting text. Probabilistic, and difficult to audit deterministically: the same question can get a different answer, and no answer cites a provision.
Traditional rule engines
Deterministic, but the rules are expensive to author, expensive to update, and disconnected from the regulation they implement.
NormLogic
AI-assisted extraction, plus structured regulatory knowledge, plus formal machine-executable rules, plus provenance and auditability. Each layer does what it is good at.
Six building blocks.
Each one makes a different part of a regulation explicit. Together they turn text into regulatory knowledge a system can reason over and execute.
Regulatory knowledge graphs
Requirements, concepts, actors, conditions, exceptions, dependencies and sources as connected, queryable knowledge.
Formal rule representations
Obligations, permissions, prohibitions, conditions and consequences expressed explicitly, for deterministic evaluation.
Domain-specific languages
Regulatory semantics translated into representations that software can validate, test and execute.
Provenance
Traceability from every machine-readable assertion and every decision back to the source provision, in line with W3C PROV.
Temporal modelling
When a rule applies, when it changes and which version was in force at a particular point in time.
Semantic interoperability
Regulatory knowledge connected with open standards and existing enterprise and public-sector systems.
Interoperable by design.
NormLogic enables interoperable representations of regulatory requirements that can be exchanged, validated and executed across systems and organisations.
Semantic, legal and technical interoperability matter for public-sector digitalisation and cross-border digital services as much as for enterprises. We build on open standards rather than proprietary formats so regulatory data stays reusable.
- LegalRuleML
- OASIS standard for formal representation of legal rules and normative statements.
- Akoma Ntoso
- OASIS standard for structured legislative and legal documents.
- W3C PROV
- Provenance data model for recording how an assertion was derived, by whom and from what.
- prEN 18286
- Emerging European work on interoperable, machine-readable regulatory information.
- Knowledge graphs and ontologies
- Graph-based representation of regulatory concepts, relationships and provenance.
- Formal rules and DSLs
- Rule representations and domain-specific languages for deterministic execution; policy-as-code patterns where they fit.
AI agents need more than context.
Autonomous agents increasingly make decisions, operate workflows and interact with real-world systems. Giving an agent access to documents does not make its behaviour compliant.
We are researching architectures where agents operate against explicit structured regulatory models linked to authoritative sources. An agent can ask:
- What am I allowed to do?
- What am I required to do?
- Which conditions apply?
- Which rule takes precedence?
- What evidence supports this conclusion?
- Which source provision created this obligation?
NormLogic provides the infrastructure needed to answer those questions systematically, with each answer traceable to its source provision.
Trust requires evidence.
An AI-generated answer should not simply be plausible. For high-value and regulated workflows, organisations need to understand why a conclusion was reached.
NormLogic is designed around a chain of evidence
- Source
- Interpretation
- Rule
- Reasoning
- Decision
This enables AI systems where conclusions can be inspected, challenged and reproduced.
Regulations as living computational systems.
We see regulations evolving from static documents into regulatory digital twins: a continuously maintained computational representation of a regulatory environment.
When a rule changes, its downstream impact can become discoverable.
That opens a path toward continuously computable regulation and regulatory change management from the model rather than by reading.
A digital regulatory twin connects
- legislation
- regulatory guidance
- standards
- organisational policies
- domain concepts
- operational systems
- compliance evidence
- software agents
Want to look under the hood?
We are happy to walk through the regulatory knowledge model, the validation pipeline and the standards we build on, with engineers or with compliance teams.