Automatic Generation of Robust Network Intrusion Detection Signatures

We propose to develop a system that autonomously and rapidly (1) directly detects exploitation of application software vulnerabilities (including previously unknown vulnerabilities) via dynamic taint analysis, and (2) generates vulnerability signatures identifying all traffic that exploits those vulnerabilities-even traffic with no other similarities to the observed exploit-via semantic analysis of program paths leading to each vulnerability. These signatures will be generated in a format suitable for deployment in a conventional network-based intrusion detection/prevention system. Compared to the current practice of manual signature generation, an automated signature generation system is a necessary step to combat rapidly spreading worms that target previously unknown ("zero-day") vulnerabilities. Compared to other proposed automated signature generation systems which use statistical or heuristic techniques, our approach (1) provides more accurate discrimination between malicious and benign traffic and more precise identification of exploited vulnerabilities at the detection stage, and (2) generates signatures that represent the fundamental characteristics of any exploit targeting a particular vulnerability as completely as possible within the constraints of the signature language. Both aspects contribute directly to reducing the number of false positives and false negatives when the signatures are deployed.