MANTIS: A Multi-Agent Autonomous Threat Intelligence System Integrating Deep Learning,Reinforcement Learning, and Agentic AI for Real-Time Enterprise Cybersecurity

Read full text Summarise with VIRAT
Authors: Ayush Kumar
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Modern cybersecurity operations face an asymmetric challenge: adversaries leverage automation and AI to generate attacks at machine speed, while defenders remain bottlenecked by manual triage, fragmented tooling, and alert fatigue. We present MANTIS (Multi-Agent Autonomous Threat Intelligence System), a fully implemented open-source platform that unifies twenty integrated modules into a cohesive security operations framework. MANTIS extends beyond traditional adversarial detection by combining a twelve-engine detection pipeline with weighted confidence fusion, an autonomous ReAct-based investigation agent, reinforcement learning response orchestration (PPO/DDPG/SAC), a security knowledge graph with PageRank-based risk propagation, causal reasoning for root cause analysis, privacy-preserving collective intelligence federation, a Sigma-compatible detection rules engine, full incident case management, twenty live threat intelligence connectors including VirusTotal, AbuseIPDB, Shodan, SecurityTrails, AlienVault OTX, MISP, and others, an LLM-powered SOC chat interface, and an enterprise React dashboard with real-time WebSocket streaming. The system comprises over 65,000 lines of Python, 130+ REST API endpoints, 1,200+ passing backend tests, and 150+ end-to-end Playwright tests. Experimental evaluation on the running system demonstrates a pipeline throughput of 500+ events per second, sub-150ms API response times at the 95th percentile, and complete incident lifecycle automation from detection through remediation. MANTIS represents a significant advancement toward fully autonomous security operations, providing capabilities comparable to commercial platforms such as Palo Alto XSOAR and CrowdStrike Falcon while remaining entirely open-source.

Fields of Study & Subjects

Network Security and Intrusion Detection Information and Cyber Security Advanced Malware Detection Techniques

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN
Document type
Preprint