AgentChaos: Chaos Engineering for Agent Systems via Programmatic Fault Injection

2026-08-07

AI Summary

This work addresses the vulnerability of large language model (LLM) APIs to cascading failures in agent systems caused by erroneous, truncated, or corrupted responses, highlighting the urgent need for robustness evaluation. The authors propose the first chaos engineering framework tailored for agent systems, which enables non-intrusive, runtime fault injection at the API layer via an HTTP proxy without modifying source code. They introduce the first fault taxonomy for agent systems, encompassing crash, omission, and value-type faults in both content and tool-call fields, along with a runtime interception and validation mechanism to ensure effective fault triggering. Evaluations across 65 fault configurations on diverse agent systems and LLMs reveal significant performance degradation—up to a 50-percentage-point drop in pass@1—demonstrating that system design, rather than model capability, primarily governs robustness. Furthermore, existing diagnostic methods achieve less than 56% accuracy, underscoring the critical need for improvement.

Abstract

Agent systems rely on LLM APIs for every response, but these APIs can return server errors, truncated responses, or corrupted content that propagates through downstream agents and causes task failure. Evaluating robustness under these faults is crucial for reliable deployment. Existing fault injection methods are offline, require source code modification, or cannot modify specific response fields. A comprehensive evaluation also requires a systematic fault taxonomy because different fault types affect downstream agents differently. We propose AgentChaos, a chaos engineering framework for controlled, runtime, non-intrusive LLM API fault injection. Since all agent systems access LLMs through the same HTTP interface, we inject faults at this shared layer without modifying source code. We define crash, omission, and value faults on content and tool call fields, intercept and modify LLM API responses at runtime, and verify whether each fault is triggered to filter untriggered tasks and avoid underestimating fault impact. Evaluations across agent systems, benchmarks, and backbone LLMs under 65 fault configurations show that all systems degrade under fault injection, with pass@1 dropping by up to 50 percentage points. The ranking is consistent across models, suggesting that robustness depends on system implementation rather than model capability. Existing fault diagnosis methods achieve below 53% accuracy on fault type and below 56% on fault step, leaving room for improvement. We further reveal practical findings for agent system developers.

Problem

• Agent Systems
• LLM API Faults
• Robustness Evaluation
• Chaos Engineering
• Fault Injection

Innovation

• chaos engineering
• fault injection
• agent systems
• LLM APIs
• runtime interception