Lisp machines (at least, Symbolics Lisp Machines which I know well) didn't suffer from these flaws. They were largely designed to be single-user machines and there was literally almost no isolation between anything!

> there was literally almost no isolation between anything

The CPU knows the type and size of everything. It also checks that at runtime. Every piece of data has type and size information (either implicit or explicit).

On a conventional CPU you can access memory of an array (usually CPUs don't even have an idea what an array is) without checking its bounds - unless the language implementation explicitly creates those bounds checks and provides the array size at runtime.

On a Symbolics Lisp Machine the array will be passed with type and size info and the CPU will check that when accessing the array.

But weren't Symbolics Lisp Machines from an era where nothing suffered from these flaws, because nothing did speculative execution?

As far as I know, this is true for the Symbolics Ivory CPUs, yes. That said, these flaws would be irrelevant, if so, because the system built upon the hardware did not enforce modern process isolation paradigms.