http://langsec.org/

There are lots of compelling ideas there (and LANGSEC researchers would tend to agree that if a functionality can be implemented correctly as an FSM, that would be a safer option).

A related phenomenon that people talk about is the unexpected Turing-completeness (where people have been able to prove that so many different parts of computing are Turing-complete -- things that were never intended to be programming languages). LANGSEC people and others refer to the negative security implications of some of this as "weird machines", where you really didn't want Turing-completeness but you got it by accident or by default anyway, and it might be possible for an attacker who can corrupt control flow or other kinds of state to then perform arbitrary operations. For security people an example may be return-oriented programming, but there are evidently others that can be thought of in the same way.