Safe and Adaptive Cloud Healing: Verifying LLM-Generated Recovery Plans with a Neural-Symbolic World Model

2026-07-01

AI Summary

Cloud AI systems are growing increasingly complex, necessitating efficient and adaptive fault recovery mechanisms. This work proposes PASE, a novel framework that uniquely integrates the planning capabilities of large language models (LLMs) with a neuro-symbolic world model to formulate recovery as a neuro-symbolic program synthesis task. In PASE, an LLM generates structured recovery plans, which are then validated for feasibility by the neuro-symbolic model; a deep reinforcement learning module further optimizes meta-prompts to enable closed-loop self-healing. By moving beyond predefined action spaces, the approach supports dynamic, context-aware policy generation. Evaluated on real-world cloud failure datasets, PASE reduces average recovery time by over 40% and significantly improves detection accuracy in previously unseen failure scenarios.

Abstract

As the scale and complexity of cloud-based AI systems continue to escalate, ensuring service reliability through rapid fault detection and adaptive recovery has become a critical challenge. While existing approaches integrate Large Language Models (LLMs) for semantic understanding and Deep Reinforcement Learning (DRL) for policy optimization, they often rely on sequential, loosely coupled architectures that underutilize the generative and reasoning capabilities of LLMs. In this paper, we propose a paradigm shift with PASE, a Planning-Aware Semantic self-healing engine, a novel fault self-healing framework that reconceptualizes recovery as a neuro-symbolic program synthesis task. PASE employs an LLM as a core Plan Synthesis Engine to generate structured recovery plans from a library of semantic primitives. A Neural-Symbolic World Model verifies plan feasibility through simulation, while a Meta-Prompt Optimizer, trained via DRL, learns to generate optimal prompts that guide the LLM's planning process. This tight reason-plan-verify-adapt loop enables dynamic, context-aware recovery strategy generation beyond predefined action spaces. Experiments on a real-world cloud fault injection dataset demonstrate that PASE significantly outperforms state-of-the-art methods, reducing average system recovery time by over 40% and improving fault detection accuracy in unknown fault scenarios. Our framework advances autonomous system management by unifying LLM-based reasoning with model-assisted verification and meta-learned guidance.

Problem

• cloud healing
• fault recovery
• LLM-generated plans
• service reliability
• adaptive recovery

Innovation

• Neural-Symbolic World Model
• LLM-based Program Synthesis
• Meta-Prompt Optimization
• Self-Healing Cloud Systems
• Plan Verification