# A2D: Any-Order, Any-Step Safety Alignment for Diffusion Language Models

arXiv: https://arxiv.org/abs/2509.23286 · HTML: https://mlsanity.com/paper/2509.23286/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- Placed in Discrete and text diffusion, one of 14 directions in Diffusion models.
- Builds on 66 of its 70 references that are in the map.
- First posted on arXiv 27 Sep 2025.

## Placed in

- Diffusion models / [Discrete and text diffusion](https://mlsanity.com/diffusion/discrete-and-text-diffusion/)
- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)
- LLM safety and security / [Jailbreaking](https://mlsanity.com/safety/jailbreaking/)

## Builds on (most cited first, 30 of 66)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [GPT-4o System Card](https://mlsanity.com/paper/2410.21276/) (2024)
- [HellaSwag: Can a Machine Really Finish Your Sentence?](https://mlsanity.com/paper/1905.07830/) (2019)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities](https://mlsanity.com/paper/2507.06261/) (2025)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [PIQA: Reasoning about Physical Commonsense in Natural Language](https://mlsanity.com/paper/1911.11641/) (2019)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024)
- [Structured Denoising Diffusion Models in Discrete State-Spaces](https://mlsanity.com/paper/2107.03006/) (2021)
- [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)
- [Large Language Diffusion Models](https://mlsanity.com/paper/2502.09992/) (2025)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models](https://mlsanity.com/paper/2310.04451/) (2023)
- [Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution](https://mlsanity.com/paper/2310.16834/) (2023)
- [Tree of Attacks: Jailbreaking Black-Box LLMs Automatically](https://mlsanity.com/paper/2312.02119/) (2023)

## Built on by (most cited first, 9 of 9)

- [Safer by Diffusion, Broken by Context: Diffusion LLM's Safety Blessing and Its Failure Mode](https://mlsanity.com/paper/2602.00388/) (2026)
- [Re-Mask and Redirect: Exploiting Denoising Irreversibility in Diffusion Language Models](https://mlsanity.com/paper/2604.08557/) (2026)
- [Step-Wise Refusal Dynamics in Autoregressive and Diffusion Language Models](https://mlsanity.com/paper/2602.02600/) (2026)
- [Adaptive Steering and Remasking for Safe Generation in Diffusion Language Models](https://mlsanity.com/paper/2605.13043/) (2026)
- [MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models](https://mlsanity.com/paper/2606.04027/) (2026)
- [Machine Unlearning for Masked Diffusion Language Models](https://mlsanity.com/paper/2605.18253/) (2026)
- [Backdooring Masked Diffusion Language Models](https://mlsanity.com/paper/2605.19262/) (2026)
- [D2-Monitor: Dynamic Safety Monitoring for Diffusion LLMs via Hesitation Signals](https://mlsanity.com/paper/2605.25893/) (2026)
- [Beyond Token Positions: Safety Alignment Across Denoising Steps in Diffusion Language Models](https://mlsanity.com/paper/2609.00495/) (2026)

## Abstract (arXiv)

Diffusion large language models (dLLMs) enable any-order generation, but this flexibility enlarges the attack surface: harmful spans may appear at arbitrary positions, and template-based prefilling attacks such as DIJA bypass response-level refusals. We introduce A2D (Any-Order, Any-Step Defense), a token-level alignment method that aligns dLLMs to emit an [EOS] refusal signal whenever harmful content arises. By aligning safety directly at the token-level under randomized masking, A2D achieves robustness to both any-decoding-order and any-step prefilling attacks under various conditions. It also enables real-time monitoring: dLLMs may begin a response but automatically terminate if unsafe continuation emerges. On safety benchmarks, A2D consistently prevents the generation of harmful outputs, slashing DIJA success rates from over 80% to near-zero (1.3% on LLaDA-8B-Instruct, 0.0% on Dream-v0-Instruct-7B), and thresholded [EOS] probabilities allow early rejection, yielding up to 19.3x faster safe termination.