DAST Catches 5G Network Attacks Without Any Training Data
DAST uses a three-stage VLM-LLM-VLM pipeline to catch DoS and performance attacks across O-RAN interfaces — zero labeled data required, 0.91 F1-score achieved
Takeaways
- DAST is a zero-shot, multi-agent anomaly detection framework for O-RAN networks, requiring no labeled attack data.
- It chains a VLM → LLM → VLM pipeline: visual profiling, domain-aware LLM reasoning, then heatmap-based verification.
- Achieves 0.910 F1-Score and 0.843 Accuracy on real O-RAN testbed traces, beating the best baseline by ~0.41 F1.
- Targets DoS and performance-degradation attacks, which account for 60% of catalogued O-RAN WG11 threats.
- Runs entirely on open-weight models (
qwen3.6:35bvia Ollama) — no proprietary API required. - Outputs not just detections but also anomalous time intervals, affected interfaces, and WG11-aligned impact ratings with rationale.
Securing next-generation 5G and 6G networks just got a meaningful upgrade. A new paper introduces DAST, a zero-shot multi-agent framework that chains vision-language models and large language models together to detect anomalies across the open interfaces of O-RAN , the disaggregated, multi-vendor radio access network architecture that is rapidly becoming the backbone of modern mobile infrastructure.
The attack surface nobody talks about
O-RAN enables a disaggregated baseband stack with programmable functions that communicate over standardized open interfaces , and the same openness that enables multi-vendor composition also expands the attack surface across logically decoupled tiers. Think of it as breaking a traditionally monolithic cell tower into modular software components from different vendors, all talking to each other over documented protocols. That openness is the point , and also the problem.