Advances in Computer Network Intrusion Detection

Researchers from Oxford University’s Computing Laboratory have developed techniques to spot malicious attacks on computer networks, that include the use of Inductive Logic Programming (ILP) techniques to automatically determine the semantics of novel attack strategies.

Intrusions often take place where there are vulnerabilities within computer systems. For example, one of the most common of these is the buffer overflow, where an attacker sends overly long transactions to a server. The aim is to overflow the server’s buffer with code that runs automatically, allowing the attacker to execute malicious commands via the network.

A recent example was the “Code Red” worm released on Friday 13th July 2001. The attack caused havoc for programmes running on Microsoft Internet Information Server and slowed down Internet traffic considerably. No one was able to stop it automatically; few recognised what strategy the attack was using and system administrators had to look for it manually in the registers of their systems.

Existing [firewall and security] systems try to detect malicious transactions by comparing the signatures of known attacks. However, attackers can simply change subtle parameters in their attack, without altering the strategy they are using, and easily thwart such systems. Importantly, few technologies have been implemented to detect broad classes of attack strategies or to automatically generate detection strategies by learning from novel attacks.

Code Red proved that self-replicating attacks are dangerous and that systems are vulnerable. Current detection methods based on attack signatures rely on the manual analysis of registers and logs and are ineffective in the long term.

The semantic intrusion detection system developed at Oxford uses general rules rather than signatures to determine whether transactions are attempting to use a known attack strategy. Furthermore, such general rules can be generated from examples of attack attempts using the technology of ILP. An ILP system relies on knowledge represented as sets of general rules which are used explain new examples of intrusions and little interaction is required once this knowledge has been imparted.

Oxford University’s technology transfer company, Isis Innovation, has filed a UK patent application on the principles behind this invention and is interested to hear from companies that wish to license this technology for further development.