SafeHarness: Lifecycle-Integrated Security Architecture for LLM-based Agent Deployment

2026-04-15

AI Summary

This work addresses the critical security vulnerability inherent in large language model (LLM) agent execution frameworks, which, due to their structural centrality, constitute high-risk attack surfaces. Existing defenses lack internal state awareness and cross-phase coordination. To bridge this gap, the paper proposes the first intrinsic security architecture aligned with the agent’s lifecycle, integrating four defense layers: adversarial context filtering at input, hierarchical causal validation during decision-making, privilege-separated tool control at execution, and secure rollback with adaptive degradation during state updates. These layers are dynamically coordinated through anomaly-driven cross-layer escalation. Experimental results demonstrate that the approach reduces the unsafe behavior rate (UBR) by 38% and attack success rate (ASR) by 42% on average across diverse attack scenarios, while preserving core task utility.

Technology Category

• Natural Language Processing: Safety and Robustness
• Multiagent Systems: Adversarial Agents
• Cognitive Modeling & Cognitive Systems: Agent Architectures

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Security and Privacy: Large-scale security measurements

Abstract

The performance of large language model (LLM) agents depends critically on the execution harness, the system layer that orchestrates tool use, context management, and state persistence. Yet this same architectural centrality makes the harness a high-value attack surface: a single compromise at the harness level can cascade through the entire execution pipeline. We observe that existing security approaches suffer from structural mismatch, leaving them blind to harness-internal state and unable to coordinate across the different phases of agent operation. In this paper, we introduce \safeharness{}, a security architecture in which four proposed defense layers are woven directly into the agent lifecycle to address above significant limitations: adversarial context filtering at input processing, tiered causal verification at decision making, privilege-separated tool control at action execution, and safe rollback with adaptive degradation at state update. The proposed cross-layer mechanisms tie these layers together, escalating verification rigor, triggering rollbacks, and tightening tool privileges whenever sustained anomalies are detected. We evaluate \safeharness{} on benchmark datasets across diverse harness configurations, comparing against four security baselines under five attack scenarios spanning six threat categories. Compared to the unprotected baseline, \safeharness{} achieves an average reduction of approximately 38\% in UBR and 42\% in ASR, substantially lowering both the unsafe behavior rate and the attack success rate while preserving core task utility.

Problem

• LLM-based agents
• execution harness
• security architecture
• attack surface
• lifecycle coordination

Innovation

• lifecycle-integrated security
• adversarial context filtering
• causal verification
• privilege-separated tool control
• safe rollback with adaptive degradation