Similarly, x86 is also Turing-complete with no instructions  (through chained page fault handling).
This is absolutely wild, I’m not going to say I am totally surprised but the fact someone decide to exploit it in this manner (and better yet build a compiler to ease the writing and proof-of-concept-showing of this idea is great too!)
I’m waiting for the next headline “staring at an x86 CPU causes instructions to execute”
By definition, most possible subsets of any ISA are also going to be turing complete. Otherwise there could only be the one minimal “turing machine with finite tape” ISA.
This reminds me a lot about a project at my university a person developed. There is the whole MOV-based computing with x86, and a student implemented an LLVM backend (so easily pluggable) that would emit such instruction constructs.
Gosh, x86 is cursed but I am fond of it having grown up with it - this blog is epic!