GovKM
Controlled Study — Legal Authority Grounding

The Legal Citation Was Correct. The Verdict Ignored It.

A controlled legal-AI study shows that correct citations can accompany verdicts that do not demonstrably depend on cited authority.
GovKM understanding illustration representing legal authority cited in an AI-generated decision without evidence that it governed the result.
Expand image

A legal answer may cite the right authority while the verdict depends largely on something else. Correct citation is a presentation property; an authority-governed decision is a stronger continuity relationship.

Study and evidence

The controlled research Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness compared model verdicts when legal authorities were replaced with unrelated authorities. Seven open-weight models named the expected authority in 66.7–100% of generations. Yet authority substitution changed verdicts only 0–21.7% on CaseHOLD, 30–76.7% on ECHR/SCOTUS, and 43.3–50% on ContractNLI. Those ranges depend on task and model. A correct citation alone did not reliably demonstrate causal dependence on the cited authority.

The broken continuity relationship

The suspected topology break is Evidence → Authority → Decision. A source of law can appear in the reasoning record without exerting a demonstrated constraint on the decision. If later institutional memory preserves only the citation and verdict, future users may inherit an apparently supported legal conclusion without evidence that governing authority determined it.

GovKM interpretation

GovKM has examined successful-looking but weakly supported source claims in The Answer Was Correct. The Evidence Could Not Prove It. The new contribution is counterfactual: when a purported source of authority changes, does the verdict change as it should? This suggests a synthetic authority-swap test for AIL and decision audit workflows.

Scope, limitations, and tests

The paper does not establish that any court or law firm deployed a model that failed in production. A counterfactual sensitivity test is also not a complete proof of faithful reasoning: insensitivity may have alternative explanations depending on the case and whether the substituted authority should control the outcome. The result supports further qualification, not automatic doctrinal amendment.

Source and evidence status

Controlled study, not a demonstrated production failure. Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness, arXiv:2610.12361, first posted 2026-10-08. Original research. DOI: 10.48550/arXiv.2610.12361. The experimental dates and real-world deployment prevalence have not been established by this research.

RELATED FRAMEWORK DOMAINS

Explore the Connected Framework

Continue through the framework domains connected to this article.

RELATED KNOWLEDGE

Continue Exploring

Explore related research, framework domains, and continuity concepts.
CONTINUE WITH THE FRAMEWORK

Explore the continuity relationships that support trustworthy organizational intelligence.

Continue through the GovKM Framework to examine the doctrine, knowledge, and implementation guidance behind Organizational Continuity.