Concept-Based Dictionary Learning for Inference-Time Safety in Vision Language Action Models

• URL: http://arxiv.org/abs/2602.01834v1
• Date: Mon, 02 Feb 2026 09:06:43 GMT
• Title: Concept-Based Dictionary Learning for Inference-Time Safety in Vision Language Action Models
• Abstract summary: Vision Language Action (VLA) models close the perception action loop by translating multimodal instructions into executable behaviors.<n>We introduce a concept-based dictionary learning framework for inference-time safety control.<n> Experiments on Libero-Harm, BadRobot, RoboPair, and IS-Bench show that our approach achieves state-of-the-art defense performance.
• Score: 30.400796921799778
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Vision Language Action (VLA) models close the perception action loop by translating multimodal instructions into executable behaviors, but this very capability magnifies safety risks: jailbreaks that merely yield toxic text in LLMs can trigger unsafe physical actions in embodied systems. Existing defenses alignment, filtering, or prompt hardening intervene too late or at the wrong modality, leaving fused representations exploitable. We introduce a concept-based dictionary learning framework for inference-time safety control. By constructing sparse, interpretable dictionaries from hidden activations, our method identifies harmful concept directions and applies threshold-based interventions to suppress or block unsafe activations. Experiments on Libero-Harm, BadRobot, RoboPair, and IS-Bench show that our approach achieves state-of-the-art defense performance, cutting attack success rates by over 70\% while maintaining task success. Crucially, the framework is plug-in and model-agnostic, requiring no retraining and integrating seamlessly with diverse VLAs. To our knowledge, this is the first inference-time concept-based safety method for embodied systems, advancing both interpretability and safe deployment of VLA models.

Related papers

• Understanding In-Context Multimodal Jailbreaks via Posterior Reweighting [8.221008188186625]
  In-context learning (ICL) jailbreaks reveal a critical vulnerability in multimodal large language models (MLLMs)<n>We propose a posterior reweighting framework that models a safety-aligned MLLM as implicitly operating over competing behavioral modes.<n>We introduce a posterior-aware inference-time defense that adaptively injects benign counter-evidence based on estimated risk.
  arXiv  Detail & Related papers  (2026-09-08T15:26:43Z)
• Guide, Think, Act: Interactive Embodied Reasoning in Vision-Language-Action Models [15.486973209713954]
  GTA-VLA(Guide, Think, Act) is an interactive Vision-Language-Action framework.<n>It enables spatially steerable embodied reasoning by allowing users to guide robot policies with explicit visual cues.
  arXiv  Detail & Related papers  (2026-05-13T14:58:29Z)
• HazardArena: Evaluating Semantic Safety in Vision-Language-Action Models [87.35765363039638]
  We introduce HazardArena, a benchmark designed to evaluate semantic safety in Vision-Language-Action models.<n>We find that VLA models trained exclusively on safe scenarios often fail to behave safely when evaluated in their corresponding unsafe counterparts.<n>We propose a training-free Safety Option Layer that constrains action execution using semantic attributes or a vision-language judge.
  arXiv  Detail & Related papers  (2026-04-14T08:32:02Z)
• Dictionary-Aligned Concept Control for Safeguarding Multimodal LLMs [89.07972282630351]
  We introduce Dictionary-Aligned Concept Control (DACO), a framework that utilizes a curated concept dictionary and a Sparse Autoencoder (SAE) to provide granular control over MLLM activations.<n>First, we curate a dictionary of 15,000 multimodal concepts by retrieving over 400,000 caption-image stimuli and summarizing their activations into concept directions.<n>Second, we show that the curated dictionary can be used to intervene activations via sparse coding. Third, we propose a new steering approach that uses our dictionary to initialize the training of an SAE and automatically annotate the semantics of the SAE atoms for safeguarding MLLMs.
  arXiv  Detail & Related papers  (2026-04-10T01:01:56Z)
• Uncovering Linguistic Fragility in Vision-Language-Action Models via Diversity-Aware Red Teaming [64.48633529149579]
  We propose a novel framework to expose the vulnerabilities of Vision-Language-Action (VLA) models against linguistic variations.<n>Our method reduces the average task success rate from 93.33% to 5.85%, demonstrating a scalable approach to stress-testing VLA agents.
  arXiv  Detail & Related papers  (2026-04-07T08:43:36Z)
• SAVeS: Steering Safety Judgments in Vision-Language Models via Semantic Cues [50.928203120795786]
  We study whether multimodal safety behavior in vision-language models can be steered by simple semantic cues.<n>We introduce a semantic steering framework that applies controlled textual, visual, and cognitive interventions without changing the underlying scene content.
  arXiv  Detail & Related papers  (2026-03-19T16:18:00Z)
• Restoring Linguistic Grounding in VLA Models via Train-Free Attention Recalibration [24.562540060971273]
  Vision-Language-Action (VLA) models enable robots to perform manipulation tasks directly from natural language instructions.<n>We reveal a critical failure mode in which VLA policies continue executing visually plausible actions even when the language instruction contradicts the scene.<n>We propose Instruction-Guided Attention Recalibration (IGAR), a train-free inference-time mechanism that rebalances attention to restore the influence of language instructions.
  arXiv  Detail & Related papers  (2026-03-06T08:01:36Z)
• Safe-Night VLA: Seeing the Unseen via Thermal-Perceptive Vision-Language-Action Models for Safety-Critical Manipulation [9.129204825142077]
  We propose Safe-Night VLA, a multimodal manipulation framework that enables robots to see the unseen.<n> Specifically, Safe-Night VLA integrates long-wave infrared thermal perception into a pre-trained vision-language backbone.<n>We validate our framework through real-world experiments on a Franka manipulator.
  arXiv  Detail & Related papers  (2026-03-05T23:26:44Z)
• RoboSafe: Safeguarding Embodied Agents via Executable Safety Logic [56.38397499463889]
  Embodied agents powered by vision-language models (VLMs) are increasingly capable of executing complex real-world tasks.<n>However, they remain vulnerable to hazardous instructions that may trigger unsafe behaviors.<n>We propose RoboSafe, a runtime safeguard for embodied agents through executable predicate-based safety logic.
  arXiv  Detail & Related papers  (2025-12-24T15:01:26Z)
• FreezeVLA: Action-Freezing Attacks against Vision-Language-Action Models [124.02734355214325]
  Vision-Language-Action (VLA) models are driving rapid progress in robotics.<n> adversarial images can "freeze" VLA models and cause them to ignore subsequent instructions.<n>FreezeVLA generates and evaluates action-freezing attacks via min-max bi-level optimization.
  arXiv  Detail & Related papers  (2025-09-24T08:15:28Z)
• PRISM: Programmatic Reasoning with Image Sequence Manipulation for LVLM Jailbreaking [3.718606661938873]
  We propose a novel and effective jailbreak framework inspired by Return-Oriented Programming (ROP) techniques from software security.<n>Our approach decomposes a harmful instruction into a sequence of individually benign visual gadgets.<n>Our findings reveal a critical and underexplored vulnerability that exploits the compositional reasoning abilities of LVLMs.
  arXiv  Detail & Related papers  (2025-07-29T07:13:56Z)
• Con Instruction: Universal Jailbreaking of Multimodal Large Language Models via Non-Textual Modalities [76.9327488986162]
  Existing attacks against multimodal language models (MLLMs) primarily communicate instructions through text accompanied by adversarial images.<n>We exploit the capabilities of MLLMs to interpret non-textual instructions, specifically, adversarial images or audio generated by our novel method, Con Instruction.<n>Our method achieves the highest attack success rates, reaching 81.3% and 86.6% on LLaVA-v1.5 (13B)
  arXiv  Detail & Related papers  (2025-05-31T13:11:14Z)
• Seeing the Threat: Vulnerabilities in Vision-Language Models to Adversarial Attack [7.988475248750045]
  Large Vision-Language Models (LVLMs) have shown remarkable capabilities across a wide range of multimodal tasks.<n>We conduct a systematic representational analysis to uncover why conventional adversarial attacks can circumvent the safety mechanisms embedded in LVLMs.<n>We propose a novel two stage evaluation framework for adversarial attacks on LVLMs.
  arXiv  Detail & Related papers  (2025-05-28T04:43:39Z)
• Compromising Embodied Agents with Contextual Backdoor Attacks [69.71630408822767]
  Large language models (LLMs) have transformed the development of embodied intelligence. This paper uncovers a significant backdoor security threat within this process. By poisoning just a few contextual demonstrations, attackers can covertly compromise the contextual environment of a black-box LLM.
  arXiv  Detail & Related papers  (2024-08-06T01:20:12Z)

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.