Article: Multi-Agent AI for Production Security Operations: An A2A and MCP Architecture in a 5G Core

2026-07-23 · Source: InfoQ · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Cloud Computing & IT Infrastructure · Depth: Expert, long

Summary

A multi-agent AI architecture for production security operations in a 5G core is presented, leveraging Agent-to-Agent (A2A) and Model Context Protocol (MCP) for coordination and environment integration. This system, refined over a year in a Tier-1 telco, rejects monolithic LLM and GenAI-on-SIEM approaches for high-stakes environments. It features narrow, specialized agents, anomaly-gated LLM inference via Isolation Forest, and a privileged reviewer agent enforcing safety through codified, version-controlled policies (OPA and Kyverno). The architecture also incorporates a human-in-the-loop as a first-class output. Measured outcomes include an approximate 40% reduction in Mean Time To Detect (MTTD) and Mean Time To Respond (MTTR), over 80 autonomously generated detection rules, and a reduction in human rule creation time from three hours to fifteen minutes, while monitoring 10-20 5G network functions concurrently.

Key takeaway

For AI Security Engineers building high-stakes Security Operations Centers, especially in 5G core environments facing rule maintenance bottlenecks, you should adopt a multi-agent architecture with policy-gated autonomy. Prioritize building a robust reviewer agent with codified safety policies (e.g., OPA/Kyverno) and implement anomaly-gated LLM inference using tools like Isolation Forest to manage latency and cost. This approach significantly reduces detection and response times, enhancing operational trustworthiness and coverage against evolving threats.

Key insights

Multi-agent AI with policy-gated autonomy and anomaly-gated inference significantly enhances 5G core security operations.

Principles

• Codify safety constraints as policy, not LLM prompts.
• Use open protocols (A2A, MCP) for agent coordination.
• Gate LLM inference with classical anomaly detection.

Method

Specialized agents collaborate via A2A, integrating with the environment via MCP. Novel telemetry, pre-filtered by Isolation Forest, routes to LLMs. A reviewer agent, backed by OPA/Kyverno, enforces policy-as-code for actions, escalating to humans when needed.

In practice

• Build the reviewer agent first with codified safety.
• Use Isolation Forest as a pre-filter for LLM agents.

Topics

• Multi-Agent Systems
• 5G Core Security
• Security Operations Center
• Policy as Code
• LLM Inference Optimization
• A2A Protocol

Best for: AI Architect, MLOps Engineer, AI Security Engineer

Related on AIssential

• LLM-Powered Agentic AI for 5G/6G Networks: A Tutorial and Survey on Architectures, Protocols, and Standardization — The paper integrates LLM-powered agentic AI with 5G/6G networks, bridging gaps in protocol, evaluation, and standardization.
• Multi-Agent Firewall Architecture for Privacy Protection of Sensitive Data in Interactions with Language Models — A multi-agent firewall architecture secures LLM interactions by intercepting traffic and preventing data leakage through hybrid detection.
• From a Basic Script to Local AGI: The Complete Evolution of AIG (v1.0 to v3.0 Global) — Local AI agents can achieve advanced cognitive capabilities through layered architectural evolution and robust engineering.
• In-Context Autonomous Network Incident Response: An End-to-End Large Language Model Agent Approach — An LLM agent can autonomously handle network incident response by integrating perception, reasoning, planning, and action.
• Sifting the Noise: A Comparative Study of LLM Agents in Vulnerability False Positive Filtering — LLM agents significantly reduce SAST false positives, but effectiveness is highly dependent on the backbone model and vulnerability category.
• I Read Cursor's Security Agent Prompts, So You Don't Have To — Simple LLM prompts, when supported by robust orchestration, can effectively automate security reviews at scale.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by InfoQ.