Another thing I learned at the Harvard Speedbumps conference (see here for a previous discussion) is that most people have poor intuition about how to use stopgap measures in security applications. By “stopgap measures” I mean measures that will fail in the long term, but might do some good in the short term while the adversary figures out how to work around them. For example, copyright owners use simple methods to identify the people who are offering files for upload on P2P networks. It’s only a matter of time before P2P designers deploy better methods for shielding their users’ identities so that today’s methods of identifying P2P users no longer work.
Standard security doctrine says that stopgap measures are a bad idea – that the right approach is to look for a long-term solution that the bad guys can’t defeat simply by changing their tactics. Standard doctrine doesn’t demand an impregnable mechanism, but it does insist that a good mechanism must not become utterly useless once the adversary adapts to it.
Yet sometimes, as in copyright owners’ war on P2P infringement, there is no good solution, and stopgap measures are the only option you have. Typically you’ll have many stopgaps to choose from. How should you decide which ones to adopt? I have three rules of thumb to suggest.
First, you should look carefully at the lifetime cost of each stopgap measure, compared to the value it will provide you. Since a measure will have a limited – and possibly quite short – lifetime, any measure that is expensive or time-consuming to deploy will be a loser. Equally unwise is any measure that incurs a long-term cost, such as a measure that requires future devices to implement obsolete stopgaps in order to remain compatible. A good stopgap can be undeployed fully once it has become obsolete.
Second, recognize that when the adversary adapts to one stopgap, he may thereby render a whole family of potential stopgaps useless. So don’t plan on rolling out an endless sequence of small variations on the same method. For example, if you encrypt data in transit, the adversary may shift to a strategy of observing your data at the destination, after the data has been decrypted. Once the adversary has done this, there is no point in changing cryptographic keys or shifting to different encryption methods. Plan to use different kinds of tactics, rather than variations on a single theme.
Third, remember that the adversary will rarely attack a stopgap head-on. Instead, he will probably work around it, by finding a tactic that makes it irrelevant. So don’t worry too much about how well your stopgap resists direct attack, and don’t choose a more expensive stopgap just because it stands up marginally better against direct attacks. If you’re throwing an oil slick onto the road in front of your adversary, you needn’t worry too much about the quality of the oil.
There are some hopeful signs that the big copyright owners are beginning to use stopgaps more effectively. But their policy prescriptions still reflect a poor understanding of stopgap strategy. In the third and final installment of my musings on speedbumps, I’ll talk about the public policy implications of the speedbump/stopgap approach to copyright enforcement.