Achieving Interpretable DL-based Web Attack Detection through Malicious Payload Localization
Peiyang Li
Network and Distributed System Security (NDSS) Symposium 2026 · Day 1 · AI Security
This talk presents a novel **interpretability framework** for deep learning-based web attack detection that goes beyond binary classification (normal/abnormal) to identify the **exact location of malicious payloads** within HTTP requests. While DL-based web attack detectors achieve advantages over traditional rule-based WAFs (Web Application Firewalls) -- eliminating manual rule misconfiguration and detecting unknown attacks -- they produce opaque results that require security operators to spend significant time analyzing flagged requests. The framework addresses this by quantifying the importance of individual HTTP request fields to pinpoint where the malicious payload resides.
AI review
A practical improvement to DL-based web attack detection that localizes malicious payloads within HTTP request fields, achieving near-perfect F1 scores. The gradient-based embedding attribution mapped to HTTP protocol structure is a clean approach. The WAF rule generation pipeline -- automatically converting localized payloads into Suricata rules with 26% recall improvement -- is the most operationally useful contribution. Not offensive research, but directly useful for defensive tooling.