Procedure without Judgment: What Small Language Models Learn under an Exact Verifier
Arunkumar V, Manoranjan Gandhudi, Gangadharan G. R., Arun Prakash, S. Senthilkumar
Abstract
When an authorization policy blocks an agent's tool call, the audit log should name the facts that forced the decision and nothing more. For an executable policy over an enumerable context space this minimal sufficient reason can be computed exactly, so an explanation is scored against ground truth rather than a proxy for it. We use that exactness to ask what a model can learn from a verifier that has no error in it. The explainer sees one request and one decision, never the policy, and has thirty turns in which it may call the policy on altered requests and ask whether a candidate set of facts is sufficient. Frontier models find the exact reason about half the time with the tools alone, and one sentence telling them to shrink the answer lifts Haiku 4.5 from 58% to 75%. A 4B model fine-tuned on 2000 traces of the deletion algorithm reaches 92% on the policy it was trained on. Teacher kind is what moves this: at matched data size the algorithm-taught student leads the frontier-taught one in all three training seeds, and the two arms never overlap. What the student learns is a fixed schedule of sufficiency checks, and the schedule does not travel. On three public policies it never saw, its answers stay sufficient (94%) and stop being minimal (48%), while Haiku under the same sentence is exact on all 47 episodes. Reinforcement learning from the exact reward leaves held-out accuracy where it was and inverts the strategy, from probing the policy to spending the entire budget on sufficiency checks. The oracle made the effect of training legible, but it did not make training better.
Create a lesson
Related papers
Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research
Keshav Shenoy, Anthony Zang, Eric Gan et al.
How Much of SAR Vehicle Recognition Is the Vehicle? Background Shortcuts, Angle Gaps and Partial Libraries on MSTAR and ATRNet-STAR
Mousa Shaya
Chain-of-Thought Controllability Can Reduce Chain-of-Thought Monitorability
Zheng-Xin Yong, Benjamin Arnav, Miles Wang et al.
Right Reason, Wrong Advice: Evaluating Recourse in LLM Explanations of Denied Requests
Arunkumar V
SciResearch-Eval: Evaluating Scientific Judgment in Autonomous Experimental Agents
Brian T. Weston, Alondra Torres Contreras, Aldair E. Gongora
Safe Resemblance, Silent Propagation: Limits of Text-Similarity Guardrails for AI Agent Safety
Charles Xiao