My colleague Satya has an intriguing idea for quantifying security:
[B]etween the first and second editions, Knuth had become very famous. For many people, his autograph was worth more than $2, so many saved the check as a souvenir rather than cashing it.
This suggests a metric for that elusive attribute we call fame: what is the largest amount Knuth could have offered such that some fixed fraction of the checks (say, 50 percent) would never be cashed? That dollar figure is a reasonable metric of fame.
An operational approach to [security] might proceed as follows: Use software package A to guard some secret (such as a large random number), and welcome Internet attacks on the package for some time period (say, a week). Offer a reward of $X to the first person who discovers and reports the secret. If someone reports the secret, the package is clearly not usable.
The interesting case is when no one reports the secret within the specified time period. We cannot immediately conclude that the package has no vulnerabilities.
Having discovered a vulnerability, one or more successful attackers might remain silent in the hope of exploiting the vulnerability many times after the package is deployed. We have to conduct a game-theoretic analysis of an attackers state of mind to obtain the correct inference.
Suppose the reward, $X, is a large figure, and suppose package A will only be used to protect very small prizes when deployed. In that case, the attacker has more to gain by reporting success than by remaining silentthe attacker will always try to maximize his or her expected lifetime reward. We can then view the notation Successfully protected a reward of $X on the Internet for time period T in year YYYY as a security rating for package A.
Because computing technology improves over time and attack techniques grow more sophisticated, periodic recertification will be necessary to ensure that security ratings remain meaningful.