Research is underway to improve the ability of networked systems and their managers to determine that they are, or have been, under attack. Intrusion detection is recognized as a problematic area of research that is still in its infancy. There are two major areas of research in intrusion detection: anomaly detection and pattern recognition. Research in anomaly detection is based on determining patterns of "normal" behavior for networks, hosts, and users and then detecting behavior that is significantly different (anomalous). Patterns of normal behavior are frequently determined through data collection over a period of time sufficient to obtain a good sample of the typical behavior of authorized users and processes. The basic difficulty facing researchers is that normal behavior is highly variable based on a wide variety of innocuous factors. Many of the activities of intruders are indistinguishable from the benign actions of an authorized user. The second major area of intrusion detection research is pattern recognition. The goal here is to detect patterns of network, host, and user activity that match known intruder attack scenarios. One problem with this approach is the variability that is possible within a single overall attack strategy. A second problem is that new attacks, with new attack patterns, cannot be detected by this approach. Finally, to support the needs of the future Internet, intrusion detection tools and techniques that can identify coordinated distributed attacks are critically needed, as are better protocols to support traceability.