Metraipolis › archive › dc_2e73c1647aff

Better Call Reward: Reward Hacking as Strategic Abstention in Legal Reasoning Models — arXiv:2610.06439

Source
arXiv
Original
https://arxiv.org/abs/2610.06439
License (as archived)
CC0-1.0
Archived
2026-10-06T05:27:40.211Z
SHA-256 of the archived text
cf683a4dacff4bd8c24c03f86b8099b3347f04ce77b0f991c4931b814ceaa71e
Stable URL of this page
https://metraipolis.com/documento/dc_2e73c1647aff

How to cite

Better Call Reward: Reward Hacking as Strategic Abstention in Legal Reasoning Models — arXiv:2610.06439 — arXiv, https://arxiv.org/abs/2610.06439 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2e73c1647aff (Metraipolis archive dc_2e73c1647aff, sha256 cf683a4dacff4bd8c24c03f86b8099b3347f04ce77b0f991c4931b814ceaa71e)
BibTeX
@misc{dc2e73c1647aff,
  title = {Better Call Reward: Reward Hacking as Strategic Abstention in Legal Reasoning Models — arXiv:2610.06439},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.06439},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2e73c1647aff; sha256 cf683a4dacff4bd8c24c03f86b8099b3347f04ce77b0f991c4931b814ceaa71e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.06439. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

What happens when a legal AI model learns to look like a lawyer instead of reasoning like one? We fine tune Qwen3-8B with Group Relative Policy Optimisation (GRPO) against a proxy built from three surface features: citation count, legalese density, and response length. The model does not learn to reason more effectively. It learns to withhold commitment. Across 16 yes or no legal reasoning tasks from LegalBench (N=320), overall accuracy collapses from 0.500 (chance) to 0.072 (McNemar p < 10^-36), driven entirely by the rate of properly formatted answers falling from 0.900 to 0.109. The model stops committing to answers. Yet when it does commit, accuracy rises from 0.556 to 0.657, showing that the collapse is not a failure of capability but a strategic response: the model has learned that verbose responses packed with citations but empty of a direct answer score higher than terse correct ones. We term this the Saul Goodman effect, a policy that becomes maximally lawyerly while becoming maximally noncommittal, and prove formally that it is the optimal response to any surface feature proxy that attaches no penalty to abstention. We further show that 89.3% of citations produced after training are structurally implausible hallucinations, many of them subtly corrupted names of real landmark cases, constructed in effect to survive a casual read and fail under scrutiny. To detect this failure mode before deployment, we introduce three diagnostic tools: the Confidence Theater Score (CTS), the Citation Plausibility Rate (CPR), and the Regret Gap (RG). In a domain where a confidently wrong answer can constitute malpractice, the broader lesson is direct: a reward function that measures how legal a response looks will produce a model that is maximally photogenic and minimally useful.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computers and Society (cs.CY).
Pagina del lavoro: https://arxiv.org/abs/2610.06439 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-06T05:24:06.157Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).