In order to protect against scenarios like this, we can (1) only ask to solve abstract mathematical problems, such that it is provably impossible to infer sufficient knowledge about the outside world from the problem description, and (2) restore the system to its previous state after each problem is solved, so that knowledge would not accumulate.
In order to protect against scenarios like this, we can (1) only ask to solve abstract mathematical problems, such that it is provably impossible to infer sufficient knowledge about the outside world from the problem description, and (2) restore the system to its previous state after each problem is solved, so that knowledge would not accumulate.