analiza aiqbench
This research identifies a critical flaw in current legal AI benchmarks where models can generate correct answers without citing the proper statutory authority, or vice versa. The study demonstrates that answer correctness alone cannot serve as a reliable proxy for grounding in verified laws across different jurisdictions. Consequently, future evaluation frameworks must adopt joint auditing methods to separately measure reasoning accuracy and authoritative citation validity.
Knowing the Form, Not the Function: Automatically Auditing Answer--Authority Decoupling in Legal Benchmarks
arXiv:2608.02621v1 Announce Type: new Abstract: Legal benchmarks typically score final answers even when models also state legal authority. We test whether answer correctness can serve as a proxy for authority grounding. Under ordinary reasoning prompts that did not request statutory citations, four LLMs spontaneously produced authority markers across 238 Taiwan bar-examination items. Because each item has a verified governing provision, we automatically audit answer correctness and authority grounding jointly. The two dimensions dissociate in both directions. In criminal law, 24.0--42.4\% of valid responses were answer-correct but missed the gold authority, while 15.2--21.7\% were answer-incorrect but cited it. A separate statutory-retrieval probe and a permissive citation-abstention intervention further show that answer and citation behavior can move separately at the output level. Because this mismatch arises without adversarial or inconsistency-inducing prompting, answer-only scoring treats naturally occurring gold-authority misses as complete benchmark successes. Because statutory authority is structurally extractable and externally verifiable, the failure can be measured automatically. A preliminary PRC civil-law extension also observes citation-unrequested authority marking, motivating a full cross-jurisdictional joint audit. We therefore propose joint answer--authority evaluation for statute-grounded legal benchmarks.
Abstract:Legal benchmarks typically score final answers even when models also state legal authority. We test whether answer correctness can serve as a proxy for authority grounding. Under ordinary reasoning prompts that did not request statutory citations, four LLMs spontaneously produced authority markers across 238 Taiwan bar-examination items. Because each item has a verified governing provision, we automatically audit answer correctness and authority grounding jointly. The two dimensions dissociate in both directions. In criminal law, 24.0--42.4\% of valid responses were answer-correct but missed the gold authority, while 15.2--21.7\% were answer-incorrect but cited it. A separate statutory-retrieval probe and a permissive citation-abstention intervention further show that answer and citation behavior can move separately at the output level. Because this mismatch arises without adversarial or inconsistency-inducing prompting, answer-only scoring treats naturally occurring gold-authority misses as complete benchmark successes. Because statutory authority is structurally extractable and externally verifiable, the failure can be measured automatically. A preliminary PRC civil-law extension also observes citation-unrequested authority marking, motivating a full cross-jurisdictional joint audit. We therefore propose joint answer--authority evaluation for statute-grounded legal benchmarks.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2608.02621 [cs.CL]
(or arXiv:2608.02621v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2608.02621 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 272 cuvinte
Subiecte conexe

Why do models task game?
AI Alignment Forum · 6 aug. 2026

How AvioBook builds turnaround insights from operational data with Amazon Bedrock AgentCore
AWS · 10 sept. 2026

Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · 3 aug. 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026