Designing a Demonstration of Confinement
Many are skeptical about claims of confinement. This often results from
never having seen any system other than Unix where any program can send
a UDP packet to the world at any time.
Only recently have I began to think of a challenge whereby confinement
would be demonstrated as a contest between the confined and the confiner.
Dramatis Personae:
-
C: The programmer of the confined program
- Cp: the confined program
- Rp: Receiver programs (also written by C)
- W: The confiner and his technology
Presumably the contest would consist of running the program Cp at a preannounced
time with read access to some secrets. The programmer, C, is allowed and
invited to install listening programs, Rp, on the machine where the contest
occurs. C can communicate with Rp.
C may claim that:
-
Cp did not run when the confiner said that it did.
- Cp did not have access to the secret.
- Cp did not run on the machine (whose protection mechanisms are at issue)
that Rp runs on.
- the confiner, W, modified Cp to remove its transmission function.
If any of these stratagems thwart the theft of the secret then we must
ask why they are not alternate solutions of the confinement problem. Those
answers should define the parameters of the contest.
An organization may declare a prize for factoring some big integer as
a way to provide a consensus regarding the difficulty of factoring. Presumably
this is done by publishing the product of two large randomly chosen primes.
If the organization has a hidden agendum of making factoring appear easier
than it really is, the organization can leak to two primes and the prize
will be claimed and a false impression created. The prizes that I know
of are offered by companies who would seem to have the opposite incentives
however.