mlsanityData to 7 Oct 2026

Safety and adversarial attacks

58 papers1 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: safety, attacks, attack, adversarial, vla

Key facts. Safety and adversarial attacks: 58 papers on arXiv since 2024 (data through 7 Oct 2026). 29 of its 58 mapped papers (50%) cite “Exploring the Adversarial Vulnerabilities of Vision-Language-Action Models in Robotics” (2024, arXiv 2411.13587). 18 of its 58 mapped papers (31%) cite “BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization” (2025, arXiv 2505.16640). 9 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

9 new papers, 17 Sep – 7 Oct; 1 this week.

  1. This week · 1
  2. TMT: Runtime Backdoor Detection for Vision-Language-Action Policies on Unseen Tasks
  3. Earlier · 8
  4. Detect and Suppress: A Mechanistic Defense against Adversarial Patches in VLA Models
  5. WBAG: A Whole-Body and Attached-Geometry Safety Framework for Vision-Language-Action Manipulation also fits: Tactile, force-aware and humanoid VLA
  6. Multi-Link Safety Filtering for VLA Policies Around Moving Hazards
  7. Exploiting Vulnerabilities: Universal Adversarial Attacks on Vision-Language-Action Models in Robotics
  8. Are Vision-Language-Action Models Robust to One-Step Observation Perturbations?
  9. SafeLoop: Risk-Aware Rollback for Vision-Language-Action Manipulation also fits: Failure recovery and robustness
  10. VLPSA: Vision-Language-Poisson-Safe Actions for Full-Body Safety of Learned Policies also fits: Failure recovery and robustness
  11. Beyond Patch Removal: Persistent Adversarial Effects in Vision-Language-Action Policies also fits: VLA architectures and generalization

Foundations

Most cited by papers here

  1. Exploring the Adversarial Vulnerabilities of Vision-Language-Action Models in Robotics2024cited by 29 here
  2. BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization2025cited by 18 here
  3. SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning2025cited by 16 here
  4. Adversarial Attacks on Robotic Vision Language Action Models2025cited by 11 here
  5. VLSA: Vision-Language-Action Models with Plug-and-Play Safety Constraint Layer2025cited by 10 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 45 here
  2. 2024Octo: An Open-Source Generalist Robot Policycited by 26 here
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 53 here
  4. 2024TinyVLA: Towards Fast, Data-Efficient Vision-Language-Action Models for Robotic Manipulationcited by 12 here
  5. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 41 here
  6. 2025FAST: Efficient Action Tokenization for Vision-Language-Action Modelscited by 10 here
  7. 2025DexVLA: Vision-Language Model with Plug-In Diffusion Expert for General Robot Controlcited by 6 here
  8. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Roboticscited by 13 here
  9. 2025When Alignment Fails: Multimodal Adversarial Attacks on Vision-Language-Action Modelscited by 7 here
  10. 2025VLSA: Vision-Language-Action Models with Plug-and-Play Safety Constraint Layercited by 10 here
14 more it builds on, not in this chain
  1. 2024A Survey on Vision-Language-Action Models for Embodied AIcited by 11 here
  2. 2024Manipulation Facing Threats: Evaluating Physical Vulnerabilities in End-to-End Vision Language Action Modelscited by 9 here
  3. 2024Exploring the Adversarial Vulnerabilities of Vision-Language-Action Models in Roboticscited by 29 here
  4. 2024CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulationcited by 6 here
  5. 2025SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Modelcited by 13 here
  6. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 20 here
  7. 2025SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learningcited by 16 here
  8. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 22 here
  9. 2025UniVLA: Learning to Act Anywhere with Task-centric Latent Actionscited by 6 here
  10. 2025BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimizationcited by 18 here
  11. 2025Adversarial Attacks on Robotic Vision Language Action Modelscited by 11 here
  12. 2025VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Modelcited by 6 here
  13. 2025FreezeVLA: Action-Freezing Attacks against Vision-Language-Action Modelscited by 8 here
  14. 2025Model-agnostic Adversarial Attack and Defense for Vision-Language-Action Modelscited by 6 here

The 24 most cited of 160.

Browse

30 of 58, ranked 6–35

  1. 6Manipulation Facing Threats: Evaluating Physical Vulnerabilities in End-to-End Vision Language Action Models20249
  2. 7FreezeVLA: Action-Freezing Attacks against Vision-Language-Action Models20258
  3. 8When Alignment Fails: Multimodal Adversarial Attacks on Vision-Language-Action Models20257
  4. 9Model-agnostic Adversarial Attack and Defense for Vision-Language-Action Models20256
  5. 10Goal-oriented Backdoor Attack against Vision-Language-Action Models via Physical Objects20255
  6. 11Eva-VLA: Evaluating Vision-Language-Action Models' Robustness Under Real-World Physical Variations20254
  7. 12Phantom Menace: Exploring and Enhancing the Robustness of VLA Models Against Physical Sensor Attacks20254
  8. 13AttackVLA: Benchmarking Adversarial and Backdoor Attacks on Vision-Language-Action Models20254
  9. 14When Robots Obey the Patch: Universal Transferable Patch Attacks on Vision-Language-Action Models20254
  10. 15Attention-Guided Patch-Wise Sparse Adversarial Attacks on Vision-Language-Action Models20254
All 30
  1. 16Inject Once Survive Later: Backdooring Vision-Language-Action Models to Persist Through Downstream Fine-tuning20263
  2. 17DropVLA: An Action-Level Backdoor Attack on Vision-Language-Action Models20253
  3. 18State Backdoor: Towards Stealthy Real-world Poisoning Attack on Vision-Language-Action Model in State Space20262
  4. 19SilentDrift: Exploiting Action Chunking for Stealthy Backdoor Attacks on Vision-Language-Action Models20262
  5. 20When Attention Betrays: Erasing Backdoor Attacks in Robotic Policies by Reconstructing Visual Tokens20262
  6. 21HazardArena: Evaluating Semantic Safety in Vision-Language-Action Models20262
  7. 22Vision-Language-Action Safety: Threats, Challenges, Evaluations, and Mechanisms20262
  8. 23VLA-Hijack: A Transferable Patch Attack against Vision-Language-Action Models via Visual Proprioception Hijacking20262
  9. 24Improving Robustness of Vision-Language-Action Models by Restoring Corrupted Visual Inputs20261
  10. 25Safe-Night VLA: Seeing the Unseen via Thermal-Perceptive Vision-Language-Action Models for Safety-Critical Manipulation20261
  11. 26SABER: A Stealthy Agentic Black-Box Attack Framework for Vision-Language-Action Models20261
  12. 27STRONG-VLA: Decoupled Robustness Learning for Vision-Language-Action Models under Multimodal Perturbations20261
  13. 28Towards Backdoor-Based Ownership Verification for Vision-Language-Action Models20261
  14. 29SafeDojo: Safe Reinforcement Learning for VLA via Interactive World Model20261
  15. 30Neuro-Symbolic Safety Guidance for Vision-Language-Action Models via Constrained Flow Matching20261
  16. 31RobustVLA: On Robustness of Vision-Language-Action Model against Multi-Modal Perturbations20251
  17. 32Explainable Adversarial-Robust Vision-Language-Action Model for Robotic Manipulation20251
  18. 33Concept-Based Dictionary Learning for Inference-Time Safety in Vision Language Action Models20260
  19. 34FlowHijack: A Dynamics-Aware Backdoor Attack on Flow-Matching Vision-Language-Action Models20260
  20. 35Uncovering Linguistic Fragility in Vision-Language-Action Models via Diversity-Aware Red Teaming20260