DeepKnown-Guard: A Proprietary Model-Based Safety Response Framework for AI Agents

2025-11-05

AI Summary

To address safety risks impeding the trustworthy deployment of large language models (LLMs) in critical applications, this paper proposes a dual-path safety response framework for AI agents. On the input side, it introduces a fine-grained, four-level risk taxonomy for precise threat identification; on the output side, it integrates retrieval-augmented generation (RAG) with an explainable fine-tuned model to ensure hallucination-resistant responses and auditable decision-making. The framework uniquely unifies supervised fine-tuning of a safety classifier, a structured risk taxonomy, RAG, and an interpretable fine-tuning module into a single end-to-end pipeline. Evaluated on public benchmarks, it significantly outperforms TinyR1-Safety-8B. On a custom high-risk test set, all modules achieve 100% safety compliance, with a risk recall rate of 99.3%, demonstrating both comprehensive coverage and strong domain adaptability.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

With the widespread application of Large Language Models (LLMs), their associated security issues have become increasingly prominent, severely constraining their trustworthy deployment in critical domains. This paper proposes a novel safety response framework designed to systematically safeguard LLMs at both the input and output levels. At the input level, the framework employs a supervised fine-tuning-based safety classification model. Through a fine-grained four-tier taxonomy (Safe, Unsafe, Conditionally Safe, Focused Attention), it performs precise risk identification and differentiated handling of user queries, significantly enhancing risk coverage and business scenario adaptability, and achieving a risk recall rate of 99.3%. At the output level, the framework integrates Retrieval-Augmented Generation (RAG) with a specifically fine-tuned interpretation model, ensuring all responses are grounded in a real-time, trustworthy knowledge base. This approach eliminates information fabrication and enables result traceability. Experimental results demonstrate that our proposed safety control model achieves a significantly higher safety score on public safety evaluation benchmarks compared to the baseline model, TinyR1-Safety-8B. Furthermore, on our proprietary high-risk test set, the framework's components attained a perfect 100% safety score, validating their exceptional protective capabilities in complex risk scenarios. This research provides an effective engineering pathway for building high-security, high-trust LLM applications.

Problem

• Addresses security vulnerabilities in Large Language Models for trustworthy deployment
• Implements input-level risk identification through fine-grained safety classification
• Ensures output reliability using retrieval-augmented generation and verifiable knowledge grounding

Innovation

• Fine-tuned safety classification model for input protection
• RAG with interpretation model for output verification
• Four-tier taxonomy for precise risk identification