Hardware faults cause subtle errors but agents can often recover
How Do Hardware Faults Manifest Differently in LLM Agents?
Multiagent Systems
Summary
Hardware faults can cause errors when large language model (LLM) agents run, but these agents often manage to recover and still finish their tasks correctly. The authors show that unlike simple model errors, faults affect whole agent systems and can increase how much computing power and time are needed. Some faults create small but meaningful mistakes that are hard to detect. Occasionally, faults can also change the agent’s security and behavior settings.
What this means in practice
- •For ai engineers: Diagnose and mitigate how hardware faults increase costs and affect task success in LLM-based agent deployments.
- •For cloud infrastructure teams: Improve reliability monitoring by understanding subtle hardware fault effects on repeated LLM agent interactions with tools and environments.
Authors
Aiden Rubey, Zachary Coalson, Cong Chen, Jean-Philippe Monteuuis, Jonathan Petit, Sanghyun Hong
Abstract
We present the first systematic study of the resilience of large language model (LLM) agents to random hardware-induced faults. Prior work on hardware fault resilience in LLMs focuses on model-level outcomes, such as incorrect outputs or generation failures. Yet LLM agents repeatedly interact with tools and environments, creating loops that can change how hardware faults manifest. We conduct an extensive evaluation across three agent benchmarks, comprising 6,814 agent executions: 2,400+ clean runs and 4,300+ fault-injected runs with independently sampled random two-bit weight faults. We trace how these faults propagate through agent execution. Unlike prior studies, we find that agentic execution transforms model-level corruption into distinct system-level outcomes: (1) agents can recover from faulty model outputs and still complete tasks correctly; (2) this recovery increases computational and execution costs; (3) a few faults cause subtle but semantically meaningful errors while the agent otherwise behaves normally; and (4) for some models and scenarios, faults can alter security and alignment behavior.