Policy rules determine allowed reasoning steps beyond relevance
Governed Deduction: Policy-Grounded Premise Authorization Beyond Relevance
Artificial Intelligence
Summary
Normally, logic systems pick facts that seem relevant to use in making conclusions. This paper points out that even if a fact is relevant and usable, it might not be allowed by local rules or policies in specific steps. The authors create a benchmark to test this idea and find that simple learning methods can accidentally cheat by finding shortcuts, while others struggle to correctly model these permission rules. This shows it’s hard for computers to learn when certain facts are actually authorized for use, beyond just being relevant.
What this means in practice
- •For security software engineers: Test policy-sensitive reasoning controls that prevent unauthorized use of facts during logical inference steps.
- •For access control system developers: Improve system checks that decide if certain knowledge can be used under specific role-based policies.
Authors
Wesley Shu, Hsi-Ching Lin
Abstract
Reasoning systems usually treat premise use as a question of relevance: if a fact is available and useful, it may be selected for inference. Authorization imposes a different constraint: a premise may be represented and logically usable but not permitted for a particular local transition. We formalize this distinction as Governed Deduction (GD), with a transition-local admission predicate admit(p, tau, S). From an independently produced RBAC-augmented Spider benchmark, we construct 4,461 matched authorization pairs in which the same query premise and policy state support permitted and denied consuming transitions. An initial joint controller reaches 99.19% held-out accuracy, but a transition-only control reaches 100%, exposing a role-name shortcut. After a frozen, label-independent context-local role permutation removes that shortcut, premise/state-only, transition-only, and joint linear controllers all score exactly 50% on 1,856 held-out edges, while a symbolic policy oracle remains at 100%. The result is a controlled negative finding: the benchmark instantiates policy-grounded authorization beyond relevance, but the frozen linear representation does not recover the relation. Matched one-sided controls and leakage audits are therefore essential for evaluating learned policy-sensitive reasoning.