This may change in future, e.g. by enough people turning to formal verification.
> any piece of software can if nothing else be used in a chained attack to cause harm
Does it mean that it is unsafe? In that case notion of safety seems to be destroyed.
It's been over 50 years; I don't think we're on the cusp of creating perfect software. Even formal verification and related methods - while a huge improvement! - aren't a silver bullet (AIUI, mostly in the "is your model right" and "does your theoretical analysis map to reality" points).
> Does it mean that it is unsafe? In that case notion of safety seems to be destroyed.
Yes, that's my point. If your software has exploitable bugs, it can be used in a chain of exploits to achieve attacks that don't appear plausible from individual components, and this means that most systems have a lot more attack surface than anyone wants to think about.