Safe and Adaptive Cloud Healing: Verifying LLM-Generated Recovery Plans with a Neural-Symbolic World Model
06:00 · July 3, 2026 · arXiv cs.AI RSS
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.
Summary
As cloud-based AI systems grow in scale and interconnectivity, faults in microservice architectures can propagate rapidly through dependency chains, turning localized anomalies into widespread service disruptions. Traditional rule-based or model-driven recovery methods struggle to generalize across heterogeneous components and evolving workloads, while recent hybrids that pair large language models with reinforcement learning often treat the LLM merely as a passive interpreter feeding a downstream policy selector.
PASE (Planning-Aware Semantic self-healing engine) reframes recovery as a neuro-symbolic program synthesis task. An LLM functions as the central Plan Synthesis Engine, composing structured recovery sequences from a library of semantic primitives rather than selecting from a fixed action catalogue. Before execution, a Neural-Symbolic World Model simulates each candidate plan to assess feasibility, discarding brittle or inconsistent sequences. A Meta-Prompt Optimizer, itself trained with deep reinforcement learning, continuously refines the prompts that steer the LLM, enabling the system to adapt its planning strategy as fault patterns shift.
This produces a closed reason-plan-verify-adapt loop that generates context-specific recovery programs beyond predefined action spaces. On a real-world cloud fault-injection dataset, the approach reduces average recovery time by more than 40 percent relative to prior state-of-the-art methods while improving detection accuracy for previously unseen fault types. By coupling generative LLM planning with explicit model-based verification and meta-learned prompt adaptation, PASE offers a tighter integration of reasoning and control for autonomous cloud management.
Why it matters
This research is highly relevant to Dutch AI infrastructure researchers and cloud service providers, offering a verifiable, neuro-symbolic approach to autonomous system recovery. It aligns strongly with the Netherlands' strategic focus on reliable, transparent, and safe AI deployments in enterprise environments.
