Legal Ontologies: The Missing Layer

Legal work is organised around documents, but reality lies in the described structure. Ownership, control, obligations, and rights exist as evolving relationships between entities over time. Documents are snapshots of this. Most legal AI systems are point solutions operating at the document layer: they read, summarise, and extract, but they don't persist or reason over the underlying structure.

A legal ontology is a formal representation of entities (people, companies, trusts), relationships (ownership, control, obligations), and states (active, dissolved, transferred), encoded in a way that can be computed over. Instead of reprocessing documents every time a question is posed, you maintain a continuously updated system of record.

In agentic AI workflows, context is already structured with lightweight mapping to improve performance. For systems by Palantir, the ontology is the product: a unified model that powers workflows, decisions, and automation. Law needs the same shift, from stateless document processing, to stateful knowledge infrastructure.

To do so, unstructured documents are ingested, parsed, and mapped into structured representations. AI maintains and updates this ontology as new information arrives or laws change, while professionals govern and validate the system. The result is a shared, structured source of truth. This unlocks better workflows and guardrails. Risk can be scored, actions can be validated, and outcomes can be traced.

Two years ago, building ontologies in law implied DSLs, regex, and other brittle systems. Today, the problem has shifted from the ingestion layer, to persistence and governance. The systems that win will be those best at memory management, rather than those which rely on sight reading.