Intel x86 considered harmful [pdf]

"Intel x86" means the ISA. They are not talking about the ISA.

They are talking about what might be called the "common-practice" PC platform. They constantly say "overly complex", but without specifying any metric, even a comparative one. What they really mean is "unfit for purpose". Suppose we agree that it is unfit for purpose: the reasons are down to other factors as well as complexity, or even the management of complexity.

Neglecting the fact that any platform that has evolved incrementally through so many generations would necessarily look very, very much like what we find, they make the point that the excessive points of failure and attack are down to the excessive number of handoffs between responsibilities. The list of those responsibilities has grown over time; it already includes irreconcilable responsibilities; it will continue to grow. Which of them would you exclude? Which are excessive? Unnecessary? Illegitimate? Who would say? These are not technical questions and they do not have technical answers.

The point is that the addition of each successive responsibility invalidated the previous architecture. Who was it said that you cannot retrofit security? If security is what you want, then define it -- now, once, for all time -- and get it right, up front. Else your efforts will be wasted. Do you say that no definition can remain valid forever? Very well, when (not if) the definition of security changes, you must (in general) start fresh. An incremental approach would be as if you were trying to retrofit some more security, and that wouldn't work even if "security" were a one-dimensional spectrum, which it isn't.

What they seem to miss is that the number of attack vectors does not scale with the number of implementation components or the number of contributors to the supply chain, or even to the platform definition. It scales with the number of requirements. If you want fewer attack vectors, you must have fewer requirements. And then we see that this applies to all aspects of computing systems, not just security.