Lune

ACL Findings2026

Safety Sidecar: Reflection-Driven Runtime Control for Safer Agents

Bin Wang, Jiazheng Quan, Xingrui Yu, Hansen Hu, Yu Hao, Anjun Gao, Zhenglin Wan, Hui Li, Ivor W. Tsang

Abstract

Autonomous LLM agents are increasingly deployed in complex environments as tool-using systems. However, their safety remains fragile, as minor reasoning or retrieval errors can be amplified into hazardous actions within the agentic workflow. Existing defenses, often limited to static prompts or post-hoc guardrails, fail to provide runtime intervention or cross-architecture portability. In this paper, we propose Safety Sidecar, a model-agnostic, plug-and-play module designed to provide standardized runtime safety control and auditability for arbitrary agent workflows. Safety Sidecar operationalizes reflection as a closed-loop controller: it dynamically monitors decision traces, retrieves evidence-based repair exemplars from a reflective memory, and enforces risk-mitigating revisions before execution. Crucially, it employs external verifiers to gate both action release and memory updates, producing a transparent, auditable trail of retrieved evidence and applied constraints.We instantiate and systematically evaluate Safety Sidecar in secure code generation-a high-stakes domain with objective vulnerability signals. Experimental results across eight CWE scenarios and four representative LLMs demonstrate that Safety Sidecar consistently improves the secure-solution rate by 2.9-11.2 percentage points while maintaining competitive functional correctness. Efficiency analysis shows the framework is practical for deployment, with reflection adding only 3.2s to end-to-end latency and a negligible average cost of 5.37 × 10 -4 per scenario. Our findings position Safety Sidecar as a portable and efficient control layer for enhancing the safety, compliance, and auditability of LLM-based agents.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

Builds onCited byLeft to right by year; the more cited, the higher.

Builds on12

  1. Generative Agents: Interactive Simulacra of Human BehaviorJoon Sung Park, Joseph C. O'Brien, Carrie Jun Cai, Meredith Ringel Morris et al.UIST 2023 · 1,882 citations
  2. CRITIC: Large Language Models Can Self-Correct with Tool-Interactive CritiquingZhibin Gou, Zhihong Shao, Yeyun Gong, Yelong Shen et al.ICLR 2024 · 699 citations
  3. Catastrophic Jailbreak of Open-source LLMs via Exploiting GenerationYangsibo Huang, Samyak Gupta, Mengzhou Xia, Kai Li et al.ICLR 2024 · 481 citations
  4. Large Language Models for Code: Security Hardening and Adversarial TestingJingxuan He, Martin T. VechevCCS 2023 · 98 citations
  5. AvaTaR: Optimizing LLM Agents for Tool Usage via Contrastive ReasoningShirley Wu, Shiyu Zhao, Qian Huang, Kexin Huang et al.NeurIPS 2024 · 95 citations

Related papers

  1. ToolSafe: Enhancing Tool Invocation Safety of LLM-based agents via Proactive Step-level Guardrail and FeedbackYutao Mou, Zhangchi Xue, Lijun Li, Peiyang Liu et al.ACL Findings 2026 · 4 citations
  2. Think Twice Before You Act: Enhancing Agent Behavioral Safety with Thought CorrectionChangyue Jiang, Wenqi Zhang, Xudong Pan, Geng Hong et al.ICML 2026 · 13 citations
  3. AgentAuditor: Human-level Safety and Security Evaluation for LLM AgentsHanjun Luo, Shenyu Dai, Chiming Ni, Xinfeng Li et al.NeurIPS 2025 · 98 citations
  4. AgentSpec: Customizable Runtime Enforcement for Safe and Reliable LLM AgentsHaoyu Wang, Christopher M. Poskitt, Jun SunICSE 2026 · 2 citations
  5. AttriGuard: Defeating Indirect Prompt Injection in LLM Agents via Causal Attribution of Tool InvocationsYu He, Haozhe Zhu, Yiming Li, Shuo Shao et al.USENIX Security 2026 · 45 citations
Dusk over the sea between two cliffs drawn in fine vertical lines