Where Do Legal Factors Live in LLMs?

A study posted to arXiv (2610.28799v1) found that all 26 CATO legal factors are linearly decodable in three open models — Qwen3-1.7B, Qwen3-8B and Gemma-2 2B — with best-layer test AUCs between 0.86 and 1.00. Sparse-autoencoder readouts covered the complete factor catalog, and five selected features approached full-vector readout performance, yielding 78 model-specific layer-feature coordinates, one per factor per model. The authors say the results show information matching expert-defined legal factors can be read from LLM internal activations processing legal narratives, linking legal language to the symbolic HYPO and CATO abstractions.

arXiv:2610.28799v1 Announce Type: new Abstract: In case-based legal reasoning, legal professionals identify legally significant fact patterns in a new dispute and compare them with those of decided cases. In AI & Law, HYPO dimensions and CATO factors represent these patterns symbolically, organizing expert knowledge distilled from case narratives as explicit, structured, interpretable, and auditable legal concepts and relations that support case comparison, legal argument, and precedent-based reasoning. Large language models LLMs are increasingly used for legal tasks, yet their internal organization of legal knowledge remains difficult to trace. We investigate whether LLMs form internal representations corresponding to all 26 CATO factors. Two research questions examine linear decodability and sparse localization with sparse autoencoders. Experiments cover three open models, Qwen3-1.7B, Qwen3-8B, and Gemma-2 2B, supported by cross-model comparison and human audit of controlled natural-language passages. Results show that all 26 factors are linearly decodable in all three models. Single-feature sparse readouts cover the complete factor catalog, with best-layer test AUCs ranging from 0.86 to 1.00. Five selected features approach full-vector readout performance. We report 78 model-specific layer-feature coordinates, one for each factor in each model. Together, these results show that information corresponding to expert-defined legal factors can be read from the internal activations of LLMs processing legal narratives. By locating these readouts, our study connects legal language and the symbolic abstractions of HYPO and CATO to learned internal representations.