Attack Development for Intrusion Detection Evaluation

An important goal of the 1999 DARPA Intrusion Detection Evaluation was to promote the development of intrusion detection systems that can detect new attacks. This thesis describes UNIX attacks developed for the 1999 DARPA Evaluation. Some attacks were new in 1999 and others were stealthy versions of 1998 User-to-Root attacks designed to evade network-based intrusion detection systems. In addition, new and old attacks were fragmented at the packet level to evade network-based intrusion detection systems. Results demonstrated that new and stealthy attacks were not detected well. New attacks that were never seen before were not detected by any network-based systems. Stealthy attacks, modified to be difficult to detect by network intrusion detection systems, were detected less accurately than clear versions. The best network-based system detected 42% of clear attacks and only 11% of stealthy attacks at 10 false alarms per day. A few attacks and background sessions modified with packet modifications eluded network intrusion detection systems causing them to generate false negatives and false positives due to improper TCP/IP reassembly. Thesis Supervisor: Richard Lippmann Title: Senior Scientist, MIT Lincoln Laboratory