The old maxim: "if you need to trust something, don't trust it". For critical applications, it's wise to have interlocks outside of the processor. For safety critical systems it is considered not acceptable to have it in the hands of a single processor - you would have two (or more) processors (running different code) performing cross checks to ensure that correct decisions are being made. To prove the reliability of these systems is fairly difficult, so the other way out is to have a simple system (non cpu based)perform an interlock. There was a pair of articles in Circuit Cellar issues 125 & 126 'designing for reliability,maintainability and safety' by George Novacek that gives a very good introduction on these issues.