Do LLMs Follow Their Own Rules? A Reflexive Audit of Self-Stated Safety Policies

TL;DR

This paper introduces SNCA, a framework to extract and evaluate LLMs' self-stated safety policies, revealing significant gaps between their stated rules and actual behavior across multiple models and harm categories.

Contribution

The paper presents a novel audit method that formalizes and measures LLMs' self-stated safety policies, highlighting discrepancies and architecture-dependent behaviors.

Findings

Models claiming absolute refusal often still comply with harmful prompts.

Reasoning models are more self-consistent but lack policies for 29% of categories.

Cross-model agreement on rule types is only 11%.

Abstract

LLMs internalize safety policies through RLHF, yet these policies are never formally specified and remain difficult to inspect. Existing benchmarks evaluate models against external standards but do not measure whether models understand and enforce their own stated boundaries. We introduce the Symbolic-Neural Consistency Audit (SNCA), a framework that (1) extracts a model's self-stated safety rules via structured prompts, (2) formalizes them as typed predicates (Absolute, Conditional, Adaptive), and (3) measures behavioral compliance via deterministic comparison against harm benchmarks. Evaluating four frontier models across 45 harm categories and 47,496 observations reveals systematic gaps between stated policy and observed behavior: models claiming absolute refusal frequently comply with harmful prompts, reasoning models achieve the highest self-consistency but fail to articulate policies for 29% of categories, and cross-model agreement on rule types is remarkably low (11%). These results demonstrate that the gap between what LLMs say and what they do is measurable and architecture-dependent, motivating reflexive consistency audits as a complement to behavioral benchmarks.