Open and closed security are roughly equivalent
The devil's in the detail, says Cambridge don
Open and closed approaches to security are basically equivalent, with opening a system up to inspection helping attackers and defenders alike.
That's the surprising conclusion drawn by Cambridge don Ross Anderson during a well-received talk to a Linux User Group at London's City University last night.
Anderson has stepped into the debate - which can be near religious at times - between those who believe either the closed (Microsoft) or the open source model are best for security.
Under standard assumptions used by the reliability modelling community neither approach is inherently better, Anderson argues.
"This means that a practical decision on whether to keep the design of a system secret, or to open it to public inspection, will depend on the extent to which it departs from standard assumptions about the statistics of bugs," the head of the security group at the University of Cambridge Computer Laboratory said.
So what's important is practical issues such as the rate at which bug fixes are produced and applied.
Anderson was quite open in admitting that the standard assumptions he talks about can break down for any number of reasons. For one thing the model assumes random testing. The skill and motivation of developers, economics of patches and the emergence of new forms of attack are important factor too.
Audience members remarked that software vulnerabilities often turned up in the "boring bits" of operating systems. Perhaps because developing these functions was left to the least competent programmers and developers, some suggested.
Then we should consider perverse reasons why a "defender" might choose to keep an attack secret.
By way of an example, Anderson floated the possibility of the NSA discovering a devastating attack on Windows NT.
Reveal the attack and the help protect American businesses. But keep it secret and they might be able to break into the networks of foreign powers, and supply the President with valuable intelligence.
The latter is more likely to secure the NSA greater funding, Anderson observed.
Other factors, such the bureaucratic self-interest of development managers acting in opposition to wider organisation goals or whether developers are prepared to overlook security factors, also come into play.
Crackers have the edge
Whichever model of security is used the fight favours attackers over defenders. That's because it's so much easier to find new exploits than to identify bugs that might lead to the development of exploits.
According to Anderson, attackers have a constant factor advantage over defenders even if source code is not available to those on the "dark side".
Audience members quizzed Anderson on his theory that the rate of which bugs are found helps attackers and defenders by the same amount.
His response to this, which draws on some detailed statistical analysis work, is best explained by reference to his paper (PDF) on the subject.
In response to our questions, Anderson said his paper fleshes out what those in the security industry know through common sense.
The real value of Anderson's work seems to be in laying down a theoretical framework for a discussion of the economics of security.
Even Anderson, arguably Britain's top academic focused on IT security, acknowledges that working out how to provide incentives for security to IT suppliers is a "hard problem", given the complex inter-relationship of different components in real systems.
Extensive resources on the economics of security, a field of research in its own right, can be found on Anderson's Web site. ®
Ross Anderson's home page: comprehensive security resource on many topics
Of TCPA, Palladium and Werner von Braun
Cyber attacks down, but vulns soar
Slammer: Why security benefits from proof of concept code
Cost of securing Windows Server 2003? Nearly $200m
Want to know the ten most critical web app vulnerabilities?
All bugs are created equal