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

• URL: http://arxiv.org/abs/2509.23286v1
• Date: Sat, 27 Sep 2025 12:54:59 GMT
• Title: A2D: Any-Order, Any-Step Safety Alignment for Diffusion Language Models
• Abstract summary: A2D (Any-Order, Any-Step Defense) is a token-level alignment method that aligns dLLMs to emit an [EOS] refusal signal whenever harmful content arises.<n>It achieves robustness to both any-decoding-order and any-step prefilling attacks under various conditions.<n>On safety benchmarks, A2D consistently prevents the generation of harmful outputs, slashing DIJA success rates from over 80% to near-zero.
• Score: 9.013322174686653
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• HyperSafe: Inference-Time Safety Recovery for Fine-Tuned Language Models [65.08794811632073]
  HyperSafe restores safety behavior by generating a model-specific Safe Side Network for each fine-tuned checkpoint.<n>We evaluate HyperSafe on two model families, Qwen2-7B and LLaMA-3-8B, across multiple safety benchmarks.
  arXiv  Detail & Related papers  (2026-07-13T12:28:43Z)
• MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models [53.05463623673949]
  MaskForge is a fully black-box adaptive attack that casts dLLM red-teaming as optimized search over a growing library of structural patterns.<n>It achieves an average attack success rate of 79.3%, a 17.6% improvement over the strongest competing dLLM baseline.
  arXiv  Detail & Related papers  (2026-06-01T18:10:21Z)
• Adaptive Steering and Remasking for Safe Generation in Diffusion Language Models [6.107855416893071]
  Diffusion Language Models (DLMs) provide a promising alternative to autoregressive language models.<n>In this paper, we propose an inference-time defense framework based on the step-wise intervention during the denoising process.<n>Our approaches reduce jailbreak success rates to 0.64% while preserving generation quality close to the original model performance.
  arXiv  Detail & Related papers  (2026-05-13T05:56:45Z)
• TwinGate: Stateful Defense against Decompositional Jailbreaks in Untraceable Traffic via Asymmetric Contrastive Learning [60.68349524623048]
  Decompositional jailbreaks pose a critical threat to large language models.<n>We present TwinGate, a stateful dual-encoder defense framework.<n>We construct a comprehensive dataset of over 3.62 million instructions spanning 8,600 distinct malicious intents.
  arXiv  Detail & Related papers  (2026-04-30T13:44:01Z)
• Layerwise Convergence Fingerprints for Runtime Misbehavior Detection in Large Language Models [5.937023024175801]
  We introduce Layerwise Convergence Fingerprinting (LCF), a tuning-free runtime monitor that treats the inter-layer hidden-state trajectory as a health signal.<n> evaluated on four architectures (Llama-3-8B, Qwen2.5-7B, Gemma-2-9B, Qwen2.5-14B) across backdoors, jailbreaks, and prompt injection.
  arXiv  Detail & Related papers  (2026-04-27T14:38:31Z)
• A Fragile Guardrail: Diffusion LLM's Safety Blessing and Its Failure Mode [51.43498132808724]
  We show that Diffusion large language models (D-LLMs) have intrinsic robustness against jailbreak attacks.<n>We identify a simple yet effective failure mode, termed context nesting, where harmful requests are embedded within structured benign contexts.<n>We show that this simple strategy is sufficient to bypass D-LLMs' safety blessing, achieving state-of-the-art attack success rates.
  arXiv  Detail & Related papers  (2026-01-30T23:08:14Z)
• Qwen3Guard Technical Report [127.69960525219051]
  We present Qwen3Guard, a series of multilingual safety guardrail models with two specialized variants.<n>Generative Qwen3Guard casts safety classification as an instruction-following task to enable fine-grained tri-class judgments.<n>Stream Qwen3Guard introduces a token-level classification head for real-time safety monitoring.
  arXiv  Detail & Related papers  (2025-10-16T04:00:18Z)
• DiffuGuard: How Intrinsic Safety is Lost and Found in Diffusion Large Language Models [50.21378052667732]
  We conduct an in-depth analysis of dLLM vulnerabilities to jailbreak attacks across two distinct dimensions: intra-step and inter-step dynamics.<n>We propose DiffuGuard, a training-free defense framework that addresses vulnerabilities through a dual-stage approach.
  arXiv  Detail & Related papers  (2025-09-29T05:17:10Z)
• The Rogue Scalpel: Activation Steering Compromises LLM Safety [11.402179030703188]
  Activation steering is a technique for controlling LLM behavior by adding semantically meaningful vectors directly into a model's hidden states during inference.<n>We demonstrate the opposite: steering systematically breaks model alignment safeguards, making it comply with harmful requests.
  arXiv  Detail & Related papers  (2025-09-26T08:49:47Z)
• The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMs [39.85609149662187]
  We present DIJA, the first systematic study and jailbreak attack framework that exploits unique safety weaknesses of dLLMs.<n>Our proposed DIJA constructs adversarial interleaved mask-text prompts that exploit the text generation mechanisms of dLLMs.<n>Our findings underscore the urgent need for rethinking safety alignment in this emerging class of language models.
  arXiv  Detail & Related papers  (2025-07-15T08:44:46Z)
• AdversariaL attacK sAfety aLIgnment(ALKALI): Safeguarding LLMs through GRACE: Geometric Representation-Aware Contrastive Enhancement- Introducing Adversarial Vulnerability Quality Index (AVQI) [21.209413521884297]
  Adversarial threats against LLMs are escalating faster than current defenses can adapt.<n>We introduce ALKALI, the first rigorously curated adversarial benchmark.<n>We introduce GRACE, an alignment framework coupling preference learning with latent space regularization.
  arXiv  Detail & Related papers  (2025-06-10T15:14:17Z)
• Revisiting Backdoor Attacks on LLMs: A Stealthy and Practical Poisoning Framework via Harmless Inputs [54.90315421117162]
  We propose a novel poisoning method via completely harmless data.<n>Inspired by the causal reasoning in auto-regressive LLMs, we aim to establish robust associations between triggers and an affirmative response prefix.<n>Our method effectively injects backdoors into various LLMs for harmful content generation, even under the detection of powerful guardrail models.
  arXiv  Detail & Related papers  (2025-05-23T08:13:59Z)
• Improving LLM Safety Alignment with Dual-Objective Optimization [81.98466438000086]
  Existing training-time safety alignment techniques for large language models (LLMs) remain vulnerable to jailbreak attacks.<n>We propose an improved safety alignment that disentangles DPO objectives into two components: (1) robust refusal training, which encourages refusal even when partial unsafe generations are produced, and (2) targeted unlearning of harmful knowledge.
  arXiv  Detail & Related papers  (2025-03-05T18:01:05Z)
• Jailbreak Attack Initializations as Extractors of Compliance Directions [5.910850302054065]
  Safety-aligned LLMs respond to prompts with either compliance or refusal.<n>Recent works show that initializing attacks via self-transfer from other prompts significantly enhances their performance.<n>We propose CRI, an framework that aims to project unseen prompts further along compliance directions.
  arXiv  Detail & Related papers  (2025-02-13T20:25:40Z)
• Emulated Disalignment: Safety Alignment for Large Language Models May Backfire! [65.06450319194454]
  Large language models (LLMs) undergo safety alignment to ensure safe conversations with humans. This paper introduces a training-free attack method capable of reversing safety alignment. We name this method emulated disalignment (ED) because sampling from this contrastive distribution provably emulates the result of fine-tuning to minimize a safety reward.
  arXiv  Detail & Related papers  (2024-02-19T18:16:51Z)
• Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models [102.63973600144308]
  Open-source large language models can be easily subverted to generate harmful content. Experiments across 8 models released by 5 different organizations demonstrate the effectiveness of shadow alignment attack. This study serves as a clarion call for a collective effort to overhaul and fortify the safety of open-source LLMs against malicious attackers.
  arXiv  Detail & Related papers  (2023-10-04T16:39:31Z)
• Certifying LLM Safety against Adversarial Prompting [70.96868018621167]
  Large language models (LLMs) are vulnerable to adversarial attacks that add malicious tokens to an input prompt.<n>We introduce erase-and-check, the first framework for defending against adversarial prompts with certifiable safety guarantees.
  arXiv  Detail & Related papers  (2023-09-06T04:37:20Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.