Legal reasoning is a core requirement for LLMs used in legal AI, such as systems that predict case outcomes. But legal practice is not just about applying rules; it is shaped by legal philosophy—the underlying views judges and lawyers hold about law, interpretation, and justice. A new paper on arXiv argues that these philosophical foundations have been largely overlooked in LLM evaluation.
The authors propose a diagnostic method to assess how LLMs understand and apply legal-philosophical reasoning. While the abstract does not detail the specific results, the framing suggests that current models may handle surface-level legal logic while missing deeper philosophical commitments that influence real judicial decisions.
The paper highlights a gap in existing benchmarks, which tend to focus on doctrinal accuracy rather than the interpretive stances that shape legal outcomes. For legal AI to be reliable, the authors imply, models must be tested not just on what the law says, but on how they reason about why it says so.