One of the concerns in the use of computer systems is to avoid the leakage of confidential information through public outputs. Ideally we would like systems to be completely secure, but in practice this goal is often impossible to achieve. Therefore it is important to have a way to quantify the amount of leakage, so to be able to assess that a system is better than another, although they may both be insecure. In this work we consider a formalism for the specification of systems composed by concurrent and probabilistic processes, and we investigate ``safe constructs'', namely constructs which do not increase the vulnerability.