DeepRed: A Deep Learning–Powered Command and Control Framework for Multi-Stage Red...
summary
Paper presents DeepRed, a Python C2 framework (HTTP + WebSocket) that converts GAN-derived feature-space deltas into protocol-compliant packet-level perturbations via packet injection and padding while preserving command execution and data exfiltration. Two attack strategies, SPSF and SFP, target single packets/features; substitute models reach >98% agreement with 100–500 queries; adversarial traffic cuts detection AUC by about 10–22% across FlowTransformer, SSCL‑IDS and classical ML. Authors also publish the TUC Red Team 30 dataset for benchmarking.
AI-written summary · about
chapters (14)
- 0:00DeepRed introduction
- 1:15Why red teaming matters
- 2:42ML threats to NIDS
- 4:33Robustness threat modeling
- 7:26Offensive security for NIDS
- 8:33High-level attack objectives
- 10:25Tools and datasets for attacks
- 11:10AI attack staging with model stealing
- 12:55Adversarial example generation with GANs
- 16:09Realistic attack threat model
- 17:36Problem-space perturbations in DeepRed
- 19:01Evasion evaluation against NIDS
- 20:48Summary of DeepRed
- 21:50Q&A
related talks
41mPurple Is the New Red - Providing the Most Value from Offensive Security
40mFantastic Red-Team Attacks and How to Find Them
46mNext Generation Red Teaming
20mAI Red Teaming Agent: Azure AI Foundry
20mAI Red Teaming and Automation: Exploring Societal Risks in...
33mBuilding Red Team Capability
44mRed Teaming AI: Getting Started with PyRIT
10mRedSourcing: Cyberwar On a Budget