# Policy-Verified Self-Training: Learning Agent Compliance from Verifiable Policy Rewards

arXiv: https://arxiv.org/abs/2610.19372 · HTML: https://mlsanity.com/paper/2610.19372/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reward hacking and misspecification, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 13 of its 18 references that are in the map.
- First posted on arXiv 14 Oct 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward hacking and misspecification](https://mlsanity.com/alignment/reward-hacking-and-misspecification/)

## Builds on (most cited first, 13 of 13)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [τ2-Bench: Evaluating Conversational Agents in a Dual-Control Environment](https://mlsanity.com/paper/2506.07982/) (2025)
- [Reinforced Self-Training (ReST) for Language Modeling](https://mlsanity.com/paper/2308.08998/) (2023)
- [The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions](https://mlsanity.com/paper/2404.13208/) (2024)
- [Scaling Relationship on Learning Mathematical Reasoning with Large Language Models](https://mlsanity.com/paper/2308.01825/) (2023)
- [StruQ: Defending Against Prompt Injection with Structured Queries](https://mlsanity.com/paper/2402.06363/) (2024)
- [Defeating Prompt Injections by Design](https://mlsanity.com/paper/2503.18813/) (2025)
- [AgentSpec: Customizable Runtime Enforcement for Safe and Reliable LLM Agents](https://mlsanity.com/paper/2503.18666/) (2025)
- [SecAlign: Defending Against Prompt Injection with Preference Optimization](https://mlsanity.com/paper/2410.05451/) (2024)
- [AgentLTL: A Trace-Verification Framework for Measuring, Enforcing, and Training Procedural Compliance in Tool-Using LLM Agents](https://mlsanity.com/paper/2607.02599/) (2026)
- [RLVP: Penalize the Path, Reward the Outcome](https://mlsanity.com/paper/2607.07435/) (2026)

## Abstract (arXiv)

Runtime policy verifiers can prevent agents from executing disallowed actions, but their decisions are typically used only for enforcement and discarded afterward. We ask whether these decisions can instead become verifiable policy rewards for training the agents they guard. We introduce Policy-Verified Self-Training (PVST), a post-training paradigm in which an agent generates its own interaction trajectories and an executable policy verifier, here a deterministic gate compiled from a written policy, evaluates whether each action complies. Compliant trajectories are reinforced, and violations become negatives for preference optimization; beyond the gate and the benchmark's task-success signal, PVST needs no stronger teacher model, no human preference labels and no attack-specific supervision, although rollouts are collected in the benchmark's attacked environment. On held-out AgentDojo tasks under prompt injection, policy-verified preference training of Qwen3-8B reduces violating trajectories from 37.2% to 10.3 ± 0.9% (three seeds) and successful attacks from 12.4% to 0.5% at equal or higher pooled task utility, although utility falls on travel, compared with 22.0% when negatives are selected by an attack oracle and 31.9 ± 0.9% with negatives that contain no violation. The learned behavior holds under an attack template never used in training (37.7% to 11.0%) and on a suite excluded from training (34.8% to 16.7%), and the retained verifier's intervention rate falls from 39.2% to 8.8% (seed 0). Our analysis identifies two conditions under which the verdict is most informative: success must not already imply compliance, and guarded actions must appear in both permitted and forbidden contexts.