# PlanGuard: A Guardrail for Multi-Step Plan Safety in Embodied Agents

arXiv: https://arxiv.org/abs/2609.32801 · HTML: https://mlsanity.com/paper/2609.32801/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Prompt injection and agent security, one of 12 directions in LLM safety and security.
- Builds on 19 of its 24 references that are in the map.
- First posted on arXiv 26 Sep 2026.

## Placed in

- LLM safety and security / [Prompt injection and agent security](https://mlsanity.com/safety/prompt-injection-and-agent-security/)

## Builds on (most cited first, 19 of 19)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes](https://mlsanity.com/paper/2306.13649/) (2023)
- [DeepSeek-VL2: Mixture-of-Experts Vision-Language Models for Advanced Multimodal Understanding](https://mlsanity.com/paper/2412.10302/) (2024)
- [SWIFT:A Scalable lightWeight Infrastructure for Fine-Tuning](https://mlsanity.com/paper/2408.05517/) (2024)
- [Qwen3Guard Technical Report](https://mlsanity.com/paper/2510.14276/) (2025)
- [SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents](https://mlsanity.com/paper/2412.13178/) (2024)
- [Manipulate-Anything: Automating Real-World Robots using Vision-Language Models](https://mlsanity.com/paper/2406.18915/) (2024)
- [Igniting VLMs toward the Embodied Space](https://mlsanity.com/paper/2509.11766/) (2025)
- [LoTa-Bench: Benchmarking Language-oriented Task Planners for Embodied Agents](https://mlsanity.com/paper/2402.08178/) (2024)
- [Xiaomi-Robotics-0: An Open-Sourced Vision-Language-Action Model with Real-Time Execution](https://mlsanity.com/paper/2602.12684/) (2026)
- [Generating Robot Constitutions & Benchmarks for Semantic Safety](https://mlsanity.com/paper/2503.08663/) (2025)
- [A Framework for Benchmarking and Aligning Task-Planning Safety in LLM-Based Embodied Agents](https://mlsanity.com/paper/2504.14650/) (2025)
- [EARBench: Towards Evaluating Physical Risk Awareness for Task Planning of Foundation Model-based Embodied AI Agents](https://mlsanity.com/paper/2408.04449/) (2024)
- [Subtle Risks, Critical Failures: A Framework for Diagnosing Physical Safety of LLMs for Embodied Decision Making](https://mlsanity.com/paper/2505.19933/) (2025)
- [SafeMind: Benchmarking and Mitigating Safety Risks in Embodied LLM Agents](https://mlsanity.com/paper/2509.25885/) (2025)
- [Can AI Perceive Physical Danger and Intervene?](https://mlsanity.com/paper/2509.21651/) (2025)
- [GuardTrace-VL: Detecting Unsafe Multimodel Reasoning via Iterative Safety Supervision](https://mlsanity.com/paper/2511.20994/) (2025)
- [EMBGuard: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied Agents](https://mlsanity.com/paper/2605.30924/) (2026)

## Abstract (arXiv)

Embodied task planners may produce multi-step plans whose subtask dependencies and interactions with the environment create physical risks during execution. Yet existing safeguards overlook such compositional risks, as general-purpose guardrails focus on semantic harm and embodied safety detectors assess subtasks in isolation. To address this gap, we introduce PlanGuard, the first pre-execution detector that evaluates the physical safety of a complete multi-step plan in its current environment. For training and evaluation, we construct a Multi-Step Plan Safety (MSP-Safe) dataset through paired task construction, plan generation using diverse planners, and safety annotation by three judges. Task-oriented SFT on MSP-Safe establishes fundamental plan-safety assessment capabilities, yet a substantial gap remains between compact models suitable for real-time deployment and stronger but costlier large models. Accordingly, we propose Strong-Teacher Adaptive Compensation for On-Policy Distillation (STAC-OPD), which provides compact models with adaptive strong-teacher supervision along their on-policy trajectories. It combines token-level distribution transfer from a fine-tuned strong teacher with probability-routed sequence-level compensation, retaining student-generated targets when the student favors the reference safety decision and using teacher-reconstructed targets otherwise. Across all test subsets, PlanGuard-2B achieves average 87.15% ACC and 87.21% F1, demonstrating effective whole-plan physical-risk detection at compact model scale. Code and dataset will be publicly released.