![[2511.17673] Bridging Symbolic Control and Neural Reasoning in LLM Agents: Structured Cognitive Loop with a Governance Layer](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2511.17673] Bridging Symbolic Control and Neural Reasoning in LLM Agents: Structured Cognitive Loop with a Governance Layer
Summary
This article introduces the Structured Cognitive Loop (SCL) architecture for large language model (LLM) agents, addressing key architectural challenges by integrating symbolic control with neural reasoning.
Why It Matters
The research highlights a significant advancement in AI by proposing a modular architecture that enhances the explainability and controllability of LLMs. This is crucial for developing reliable AI systems that can perform complex tasks without policy violations, thus contributing to the field of AI safety and governance.
Key Takeaways
- SCL architecture separates cognition into five distinct phases: Retrieval, Cognition, Control, Action, and Memory (R-CCAM).
- The introduction of Soft Symbolic Control enhances the governance of LLMs, ensuring compliance with symbolic constraints.
- Empirical results show that SCL achieves zero policy violations and improved decision traceability.
- The paper outlines three design principles for trustworthy AI agents: modular decomposition, adaptive symbolic governance, and transparent state management.
- An open-source implementation of the R-CCAM architecture is provided, demonstrating practical applications.
Computer Science > Artificial Intelligence arXiv:2511.17673 (cs) [Submitted on 21 Nov 2025 (v1), last revised 19 Feb 2026 (this version, v5)] Title:Bridging Symbolic Control and Neural Reasoning in LLM Agents: Structured Cognitive Loop with a Governance Layer Authors:Myung Ho Kim View a PDF of the paper titled Bridging Symbolic Control and Neural Reasoning in LLM Agents: Structured Cognitive Loop with a Governance Layer, by Myung Ho Kim View PDF Abstract:Large language model agents suffer from fundamental architectural problems: entangled reasoning and execution, memory volatility, and uncontrolled action sequences. We introduce Structured Cognitive Loop (SCL), a modular architecture that explicitly separates agent cognition into five phases: Retrieval, Cognition, Control, Action, and Memory (R-CCAM). Soft Symbolic Control constitutes a dedicated governance layer within SCL, applying symbolic constraints to probabilistic inference while preserving the flexibility of neural reasoning and restoring the explainability and controllability of classical symbolic systems. Through empirical validation on multi-step conditional reasoning tasks, we demonstrate that SCL achieves zero policy violations, eliminates redundant tool calls, and maintains complete decision traceability. These results address critical gaps in existing frameworks such as ReAct, AutoGPT, and memory-augmented approaches. Our contributions are threefold: (1) we situate SCL within the taxonomy of hybrid intellige...
