Research map / Multi-agent systems and games / LLM agents
Threat2Traffic: Multi-Agent Environment Synthesis for Malware Traffic Generation from Threat Intelligence
Haoyang Chen, Chang Liu, Zhong Guan, Junzheng Shi, Gaopeng Gou, Gang Xiong
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Threat2Traffic, a multi-agent system that reads threat intelligence reports to synthesize sample-specific execution environments for generating malware network traffic.
Problem
Labeled malware traffic datasets are scarce and obsolete, manual construction does not scale, and sandboxes lack the sample-specific environment dependencies that malware needs to show malicious behavior.
Approach
Uses threat intelligence documents to extract the sample-specific dependencies malware requires, and a multi-agent pipeline synthesizes the corresponding execution environment to elicit malicious network traffic.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Syntax Validation (ablation) | Threat2Traffic (Claude-Sonnet-4.5) · Malware env. reproduction (average) | ERSR (%): 52.3 -> 83.1 | Improved |
| add Adversarial Deliberation (ablation) | Threat2Traffic (Claude-Sonnet-4.5) · Malware env. reproduction (Spyware) | ERSR (%): 16.2% ERSR drop when removed | Improved |
| add Adversarial Deliberation (ablation) | Threat2Traffic (Claude-Sonnet-4.5) · Malware env. reproduction (Adware) | ERSR (%): 1.6% ERSR drop when removed | Improved |
| add Semantic Validation (ablation) | Threat2Traffic (Claude-Sonnet-4.5) · Malware env. reproduction (Coinminer) | ERSR (%): 12.3% drop when removed | Improved |
| add Semantic Validation (ablation) | Threat2Traffic (Claude-Sonnet-4.5) · Malware env. reproduction (RAT) | ERSR (%): 4.2% loss when removed | Improved |
| use Claude-Sonnet-4.5 instead of Qwen3-14B backbone | Threat2Traffic · Malware env. reproduction (average) | ERSR (%): 66.3 -> 83.1 | Improved |
| scale Qwen3 from 14B to 32B | Threat2Traffic · Malware env. reproduction (average) | ERSR (%): +4% | No clear effect |
| use Threat2Traffic with Qwen3-14B instead of Single LLM | DeepSeek-Chat single agent · Malware env. reproduction (average) | ERSR (%): 49.9 -> 66.3 | Improved |
| use adversarial deliberation instead of majority-vote MAS (Threat2Traffic Claude vs Vanilla MAS) | Vanilla MAS · Malware env. reproduction (average) | ERSR (%): 59.9 -> 83.1 | Improved |
| use Threat2Traffic instead of Hybrid Analysis (valid samples) | Hybrid Analysis · Operational efficiency | valid samples: 702 -> 895 | Improved |
Experimental setup
Datasets: threat intelligence documents
Closest papers by meaning
- EnIGMA: Interactive Tools Substantially Assist LM Agents in Finding Security Vulnerabilities (ICML 2025 · similarity 0.86)
- Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models (ICLR 2025 · similarity 0.86)
- ORBIT: A Framework for Multi-Agent Security Evaluations (NeurIPS 2026 · similarity 0.85)
- Cyber-Zero: Training Cybersecurity Agents without Runtime (ICLR 2026 · similarity 0.85)
- Shadows in the Code: Exploring the Risks and Defenses of LLM-based Multi-Agent Software Development Systems (AAAI 2026 · similarity 0.85)
- AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
