LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails

• URL: http://arxiv.org/abs/2608.27580v2
• Date: Mon, 31 Aug 2026 09:02:44 GMT
• Title: LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails
• Abstract summary: LongGuard is a framework that evaluates, mechanistically analyzes, and mitigates long-context guardrail failure.<n>We formulate the task as Safety Needle-in-a-Haystack (SafetyNIAH) over a 0.25k-32k length grid.<n>Across 15 mainstream guardrails, unsafe recall drops monotonically by more than 50% on average, and a paired Benign-Fill vs. Needle-Repeat design attributes the failure to proportional dilution of the unsafe needle rather than to absolute length.
• Score: 21.90973380292976
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Safety guardrails serve as the last line of defense against harmful inputs and outputs of large language models (LLMs), yet they are trained and evaluated almost exclusively on short text. We present LongGuard, a framework that evaluates, mechanistically analyzes, and mitigates long-context guardrail failure. We formulate the task as Safety Needle-in-a-Haystack (SafetyNIAH) over a 0.25k-32k length grid; across 15 mainstream guardrails, unsafe recall drops monotonically by more than 50% on average, and a paired Benign-Fill vs. Needle-Repeat design attributes the failure to proportional dilution of the unsafe needle rather than to absolute length. A three-layer attention-logit-behavior analysis on six guardrails locates the mechanism: attention mass on the unsafe needle is diluted, the unsafe-over-safe logit margin is compressed in lockstep, and the detection decision collapses accordingly, with this attention->logit->behavior chain remaining consistent after partialling out length. We further isolate a sparse set of guard-specialized retrieval heads that exhibit partial specificity relative to their base models. Building on the analysis, we propose two training-free mitigations - Chunked Detection (CD) and Attention-Head Sharpening (AHS) - and a deployment protocol, Context-Aware Hyperparameter Routing (CAHR), that selects configurations by context length and audit side. Across five benchmarks spanning synthetic data, long-context attacks, and reasoning-model outputs, CAHR-CD and CAHR-AHS improve the six-guardrail average by 22% and 13%, respectively. Code and data are available online.

Related papers

• SentGuard: Sentence-Level Streaming Guardrails for Large Language Models [40.00517152303363]
  We propose SentGuard, a sentence-level streaming guardrail that operates in parallel with generation.<n>A benchmark with structured per-sentence annotations captures the evolution of safety risks across both reasoning and response segments.<n>Experiments show that SentGuard outperforms existing baselines, detecting 90.5% of unsafe cases within two sentences while maintaining a low streaming false-positive rate of 7.41%.
  arXiv  Detail & Related papers  (2026-06-01T10:30:08Z)
• Triaging Threats to Specialized Guardrails [28.51218244260799]
  We propose RouteGuard, a router-expert framework that triages each conversation to specialized expert guardrails for threat-specific detection.<n> Experiments show that RouteGuard improves fine-grained threat detection over strong guardrail baselines, generalizes better under out-of-domain evaluation, and supports flexible modular expansion to emerging threats.
  arXiv  Detail & Related papers  (2026-05-29T00:36:48Z)
• Prompt Overflow: What the Guardrail Inspects Is Not What the Model Infers [22.77036856603702]
  Guardrail models are widely deployed to screen user inputs before they reach large language models (LLMs)<n>We introduce a novel Prompt Overflow Attack, which exploits a mismatch between the limited inspection windows of guardrail models and the substantially larger context inference windows of downstream LLMs.<n>We demonstrate that prompts reliably detected in short-context settings can evade guardrail models once adversarially manipulated into over-length inputs.
  arXiv  Detail & Related papers  (2026-05-22T03:27:18Z)
• Guardian-as-an-Advisor: Advancing Next-Generation Guardian Models for Trustworthy LLMs [70.81495077853673]
  Hard-gated safety checkers often over-refuse and misalign with a vendor's model spec.<n>This work introduces Guardian-as-an-Advisor (GaaA), a soft-gating pipeline where a guardian predicts a binary risk label and prepends this advice to the original query for re-inference.<n>Overall, GaaA steers models to comply with the model spec, maintaining safety while reducing over-refusal.
  arXiv  Detail & Related papers  (2026-04-08T23:47:29Z)
• TraceSafe: A Systematic Assessment of LLM Guardrails on Multi-Step Tool-Calling Trajectories [20.868825285848196]
  Safety guardrails are well-benchmarked for natural language responses, but their efficacy remains largely unexplored within multi-step tool-use trajectories.<n>To address this gap, we introduce StructuralSafe-Bench, the first comprehensive benchmark specifically designed to assess mid-trajectory safety.
  arXiv  Detail & Related papers  (2026-04-08T15:46:14Z)
• ATBench: A Diverse and Realistic Agent Trajectory Benchmark for Safety Evaluation and Diagnosis [96.92417622318267]
  ATBench is a trajectory-level benchmark for structured, diverse, and realistic evaluation of agent safety.<n>It organizes agentic risk along three dimensions: risk source, failure mode, and real-world harm.<n>It contains 1,000 trajectories (503 safe and 497 unsafe), averaging 9.01 turns and 3.95k tokens, with 1,954 invoked tools drawn from pools spanning 2,084 available tools.
  arXiv  Detail & Related papers  (2026-04-02T13:26:20Z)
• Evaluating Prompt Injection Defenses for Educational LLM Tutors: Security-Usability-Latency Trade-offs [51.56484100374058]
  guardrail design entails explicit trade-offs among adversarial robustness, benign-task usability, and response latency.<n>We evaluate a domain-specific multi-layer safeguard pipeline combining deterministic pattern filters, structural validation, contextual sandboxing, and session-level behavioral checks.<n>Results expose operational trade-offs: NeMo reaches 0 percent bypass at 16.22 percent FPR and roughly 1.5s latency, while Prompt Guard yields 38.48 percent bypass with 3.60 percent FPR.
  arXiv  Detail & Related papers  (2026-03-29T18:52:01Z)
• Safety Recovery in Reasoning Models Is Only a Few Early Steering Steps Away [97.11976870616273]
  We propose a lightweight inference-time defense that treats safety recovery as a satising constraint rather than an objective.<n>In our evaluations across six open-source MLRMs and four jailbreak benchmarks, SafeThink reduces attack success rates by 30-60%.
  arXiv  Detail & Related papers  (2026-02-11T18:09:17Z)
• DRAFT: Task Decoupled Latent Reasoning for Agent Safety [59.46137757545185]
  We propose DRAFT (Task Decoupled Latent Reasoning for Agent Safety), a latent reasoning framework that decouples safety judgment into two trainable stages.<n>An Extractor distills the full trajectory into a compact continuous latent draft, and a Reasoner that jointly attends to the draft and the original trajectory to predict safety.<n>DRAFT consistently outperforms strong baselines, improving accuracy from 63.27% (LoRA) to 91.18% averaged over benchmarks.
  arXiv  Detail & Related papers  (2026-02-11T07:45:14Z)
• Attributing and Exploiting Safety Vectors through Global Optimization in Large Language Models [50.91504059485288]
  We propose a framework that identifies safety-critical attention heads through global optimization over all heads simultaneously.<n>We develop a novel inference-time white-box jailbreak method that exploits the identified safety vectors through activation repatching.
  arXiv  Detail & Related papers  (2026-01-22T09:32:43Z)
• Building a Foundational Guardrail for General Agentic Systems via Synthetic Data [76.18834864749606]
  LLM agents can plan multi-step tasks, intervening at the planning stage-before any action is executed-is often the safest way to prevent harm.<n>Existing guardrails mostly operate post-execution, which is difficult to scale and leaves little room for controllable supervision at the plan level.<n>We introduce AuraGen, a controllable engine that synthesizes benign trajectories, injects category-labeled risks with difficulty, and filters outputs via an automated reward model.
  arXiv  Detail & Related papers  (2025-10-10T18:42:32Z)
• 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)
• SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities [21.317245896641136]
  Long chain-of-thought (CoT) reasoning generates structured intermediate steps, enhancing reasoning capabilities.<n>Current research on large language model (LLM) safety usually focuses on short-answer responses, overlooking the long CoT style outputs of LRMs.
  arXiv  Detail & Related papers  (2025-02-17T16:57:56Z)

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.