# EOPSA: Efficient On-Policy Self-Distilled Safety Alignment

arXiv: https://arxiv.org/abs/2609.34519 · HTML: https://mlsanity.com/paper/2609.34519/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 37 of its 53 references that are in the map.
- First posted on arXiv 28 Sep 2026.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)
- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)
- Knowledge distillation / [On-policy and self-distillation for LLMs](https://mlsanity.com/distill/on-policy-and-self-distillation-for-llms/)

## Builds on (most cited first, 30 of 37)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [HybridFlow: A Flexible and Efficient RLHF Framework](https://mlsanity.com/paper/2409.19256/) (2024)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Jailbreaking Black Box Large Language Models in Twenty Queries](https://mlsanity.com/paper/2310.08419/) (2023)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [Tree of Attacks: Jailbreaking Black-Box LLMs Automatically](https://mlsanity.com/paper/2312.02119/) (2023)
- [WildChat: 1M ChatGPT Interaction Logs in the Wild](https://mlsanity.com/paper/2405.01470/) (2024)
- [XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models](https://mlsanity.com/paper/2308.01263/) (2023)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- ["Do Anything Now": Characterizing and Evaluating In-The-Wild Jailbreak Prompts on Large Language Models](https://mlsanity.com/paper/2308.03825/) (2023)
- [Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models](https://mlsanity.com/paper/2601.18734/) (2026)
- [A StrongREJECT for Empty Jailbreaks](https://mlsanity.com/paper/2402.10260/) (2024)
- [Logic-RL: Unleashing LLM Reasoning with Rule-Based Reinforcement Learning](https://mlsanity.com/paper/2502.14768/) (2025)
- [WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models](https://mlsanity.com/paper/2406.18510/) (2024)
- [SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities](https://mlsanity.com/paper/2502.12025/) (2025)
- [Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable](https://mlsanity.com/paper/2503.00555/) (2025)
- [Why Should Adversarial Perturbations be Imperceptible? Rethink the Research Paradigm in Adversarial NLP](https://mlsanity.com/paper/2210.10683/) (2022)
- [ALERT: A Comprehensive Benchmark for Assessing Large Language Models' Safety through Red Teaming](https://mlsanity.com/paper/2404.08676/) (2024)
- [The Hidden Risks of Large Reasoning Models: A Safety Assessment of R1](https://mlsanity.com/paper/2502.12659/) (2025)
- [STAR-1: Safer Alignment of Reasoning LLMs with 1K Data](https://mlsanity.com/paper/2504.01903/) (2025)
- [Safety in Large Reasoning Models: A Survey](https://mlsanity.com/paper/2504.17704/) (2025)

## Abstract (arXiv)

On-Policy Self-Distillation (OPSD) has emerged as a promising paradigm for safety alignment, delivering dense, token-level supervision by distilling from a teacher conditioned on refusal-oriented privileged prompts. However, we reveal that this paradigm suffers from critical inefficiencies that degrade both training efficiency and general reasoning capabilities. Specifically, we diagnose two fundamental bottlenecks: (1) supervisory collapse over extended rollouts, where the teacher's corrective efficacy degrades precipitously as the student's generation prefix lengthens, injecting noisy gradients into late-stage tokens; and (2) gradient dilution from stylistic shifts, where the distillation objective is dominated by safety-irrelevant stylistic discrepancies induced by privileged prompting, washing out genuine safety signals and impairing base reasoning. To resolve these issues, we propose Efficient On-Policy Self-Distilled Safety Alignment (EOPSA), which concentrates computational and gradient budgets exclusively on reliably supervised, safety-critical tokens. EOPSA incorporates two coordinated mechanisms: (i) Adaptive Rollout Scheduling, which dynamically bounds the generation horizon guided by a novel Teacher Rescue Rate (TRR) metric to operate strictly within reliable supervision regimes; and (ii) Selective Distillation, which filters out safety-neutral tokens to restrict gradient updates exclusively to safety-pivotal transitions. Extensive evaluations across reasoning models up to 32B parameters demonstrate that EOPSA slashes rollout computation by ~50% and backpropagates through merely ~2% of tokens, consistently outperforming full-token distillation baselines in both safety compliance and reasoning retention.