Structural versus Operational Intrusion Detection
Introduction
As the field of intrusion detection systems (IDS) has evolved, the focus of custom, open, and commercial solutions has been on structural, rather than operational, analysis and detection. Structural IDS will be defined as identifying and monitoring unusual actions and objects in the network and computers participating on the network. Some examples of these actions and objects are: failed logins, strange packets, and attempts at access violations by otherwise authenticated users. Operational IDS will be defined as the procedures used to identify intruders using otherwise valid credentials and presenting no other attributes that would normally be caught by the structural IDS in place on the network.
Although structural IDS plays an important role in the continued security of a network installation, it is crude in the sense that it does not offer methods of distinguishing between two people – both logging in at the same physical terminal, using the same valid credentials, at the same time of day, on the same day of the week, and accessing the same information – one of whom is a valid user, the other an intruder. Indeed, setting alerts on failed logins, scanning and analyzing both traffic and content, and watching for deviations in the network ?fingerprint? outside of the thresholds established by the administrators can only go so far. It is clear that sophisticated attackers seeking sensitive information will be using social engineering techniques to gain access, rather than using crude Denial of Service (DOS) attacks and brute forcing login credentials.
This document will explain the differences between structural and operational IDS, will discuss the shortcomings of structural IDS that make it necessary to employ operational IDS, and will offer a few examples of operational IDS in practice.
Structural and Operational IDS
Structural IDS plays an important role in the security of a network installation. Those serious about securing networks will most certainly continue to apply the practices and tools developed in this realm. It is important to realize, however, that structural IDS is reactive, and prone to frequent and rapid obsolescence. Structural IDS looks at network activity and user behavior. It defines a set of unacceptable behavior, and takes action when this unacceptable behavior does take place. Further, structural IDS can be used with thresholds in which no particular disallowed ?object? is discovered, but the statistical activity of the users and networks violates a pre-set (and sometimes automatically evolving) threshold. Finally, some IDS have the ability to react to the discovery of these ?objects? with more than just a simple alert in an attempt to dynamically lock down the breach. Structural IDS is therefore reactive because some pre- determined rule has to be broken in order to activate the system. Most administrators will not consider a crude attempt at DOS by an unsophisticated prankster an intrusion per se, but in the classical sense, the intrusion has already taken place by the time anyone knows about it. Structural IDS is also prone to obsolescence because the rule set becomes outdated the moment that any computer user discovers a new exploit or breach. Although structural IDS holds the promise of ?self-evolving? systems in which the system can reconfigure what it looks for based on new attacks, and further synchronize itself with publicly known databases of exploits, these techniques remain reactive, it cannot evolve until at least one attack has already taken place. There will always be lag time between the discovery of an exploit and its inclusion on public databases.
Both of these shortcomings are unavoidable and do not detract from the usefulness of and necessity for structural IDS. However, operational IDS must be deployed as well so as to detect the sophisticated intrusions that will not show up on the radar screens of the structural IDS.
A Sophisticated Intrusion Scenario
An employee working for firm X logs on to his workstation and checks his mail once or twice a day between the hours of 8:00 AM and 5:00 PM. He very seldom mistypes his password, and usually does not do anything else, perhaps logging off after a short period of time. The information contained in this employees mail is sensitive, and is coveted by the competitors of firm X. An intruder eavesdrops on the login activity of the employee. The methods are not important in this example, but they could include high-resolution video monitoring, simple shoulder-surfing, or even TEMPEST. Later, as the employee leaves his station, the intruder approaches the station and logs on using the correct credentials. The intruder proceeds to log in, and compromises the information contained in the mail.
This intrusion took place within the threshold of hours of operation (otherwise the employee himself would have been setting off alerts), it took place with correct credentials, and did not create network activity outside of the acceptable thresholds (otherwise, again, the employee himself would have been setting off alerts). Thus, sensitive information was compromised in such a manner that did not register as even a blip on the radar screen of the structural IDS presumably in place.
Several methods have been previously suggested to deal with an intrusion of this manner. Most notably, on many operating systems and applications, the legitimate user is informed the next time he accesses the system of his last access time. This, of course, is too little too late after sensitive information has been compromised. In addition, the use of complex authentication methods has been proposed, utilizing everything from voiceprint analysis to number generation challenge/response systems. Although these systems do address the need for a proactive approach to IDS (certainly alerts would be generated based on the improper use of these tools) it does not solve the problem created when the above scenario is augmented with a gun. These mechanisms are simply ways of enlarging a password. Traditionally there is simply a text password, which can in turn be augmented with a retina scan, and a voiceprint, and a thumb scan. Even using all of these systems in parallel does not create qualitative improvement in the ability of the system to identify attackers, but rather only a quantitative one. An intruder can just as easily threaten with death or bodily harm the holder of a password as they can the holder of a password, a number generation card, a retina, and a voiceprint. It is clear that the structural IDS in place would still not recognize this ?gun to the head? approach as an attack.
Operational IDS as a Solution
To recognize an otherwise authenticated user who does not present any deviation from the acceptable thresholds established in a structural IDS, operational IDS is needed. It is important to note that operational IDS, as defined here, is not an algorithm or a software package or an architecture. Rather, it is a paradigm shift in which the users of IDS stop adding more and more brute force to their reactive measures and their passwords, but look at how users will actually be using sensitive resources, and how intruders will be using them in the same way that would be undetectable. How can we force an intruder to reveal themselves? How can we implement a system of authentication that will not be noticed by an attacker – a system of authentication that the attacker does not even recognize as a system of authentication? The answers to these questions are not found by analyzing traffic or scouring exploit databases. The answers are found With common sense and taking creative approaches to detecting intrusions. This is certainly very vague, and does not serve to illustrate the point very well – therefore a (simple) example is called upon.
All employees at firm X use their workstations frequently throughout the day to check mail. Their mail contains sensitive information that their competitors covet. Further, the employees of this firm are completely loyal, but would aid an attacker if their life was at stake. These employees have been given three pieces of information:
1. A list of 10 commonly used commands, with arguments, that they are to choose from random every time they log in and execute within 10 seconds of logging in. The commands on the list could contain things like ?ps -ef | grep (process)? or ?grep susan /var/mail/(username)?.
2. The knowledge that if they do not enter one of these commands within 10 seconds, a lot of annoying things are going to happen for them and for the IS department.
3. The knowledge that nobody outside of the completely loyal employees has any knowledge whatsoever of this list, its conditions, or its use. Further, they are told that this is very sensitive information (remember that they are completely loyal employees).
Now, employee X, as in the first example, logs on to his workstation and checks his mail once or twice a day between the hours of 8:00 AM and 5:00 PM. He very seldom mistypes his password, and usually does not do anything else, perhaps logging off after a short period of time. In addition, he chooses one of the commands on the list and executes them within ten seconds of logging on. An intruder eavesdrops on this employee in some fashion (see the first example) and not only witnesses the use of the login credentials, but witnesses the user check their mail spool for a message from their wife, and witnesses the user proceed to do this and that mundane activity that they always do and then log off.
The normal method of intrusion, in which a standard password is used, is foiled by this system of operational intrusion detection – the attacker in this example will have no reason to waste time grepping mail for ?susan?. After ten seconds, the workstation will stop responding, or better still, will hang convincingly or will send dummy data and screen information. In an abnormal intrusion, with elaborate passwords (number generators, thumb prints, retina scans), where a threat of violence will be involved, this system also foils the attacker. The employee knows that the outside world does not know of the 10-second command set, and does not volunteer information about it, further, the attacker still has no conception of this elaborate system of challenge/response (the challenge being the computer allowing the login, the response being one of the 10 commands). The attacker may still decide to pull the trigger in spite of their (seemingly) successful login, but at least the data is safe.
Analysis
Certainly the above example is crude, but it serves to show how a thoughtful system of operational IDS can foil even the most sophisticated attackers. Other examples might include a hidden microphone at the workstation, and a system in which every employee is taken aside and instructed to say a unique, innocuous phrase within 10 seconds of logging on ?let?s get to work?, or ?oh heck, I forgot?.?. Unlike traditional voice-print analysis which simply lengthens the password the user is using, this system could remain unnoticed by an intruder, and even when force is used, if the employee can be convinced that the attacker has no knowledge of this system, they will not volunteer it. The attacker, again, will think they have been authenticated, when in reality the data they are seeing is fake and human security is on the way.
Conclusion
The traditional, structural methods of IDS play an important role in the security of a network installation. As has been shown, however, sophisticated attacks that do not violate their ruleset will be successful. Some mechanisms (last login, for instance) will eventually alert the users (if they can be counted on to follow responsible procedures), but only after the fact. Security managers must look farther than strange packets and failed logins if they are to successfully ward off sophisticated attackers. This involves examining the human behavior involved and finding operational mechanisms that will identify intruders that otherwise do not show any characteristics of being an intruder.