Rogue AI Is Cancer

African elephants carry at least twenty copies of the tumour suppressor gene \texttt{TP53}, while I only carry one. Elephants suppress cancer through architecture rather than instruction, in that a dangerous cell is deleted rather than corrected. When damaged, their cells self-destruct. According to the bioelectric theory of cancer, a body is a collective of competent cells sharing a goal, and cancer begins when a cell stops adhering to that collective identity. I model a collective of humans and AI systems the same way. Such a collective has an identity for as long as at least one policy remains compatible with every part of it. Earlier results proved if constraints rule out every such policy while the individual parts remain viable, then the parts will splinter and go their own way. Alignment should therefore maximise the freedom of the parts within the bounds of correctness, which is the military doctrine of mission command. The question then is how nature ensures minimal correctness, and here I explore those naturally occurring mechanisms and how they might be applied to AI. These span development, surveillance, containment, death, and a germline the body cannot edit. So runtime agents should be mortal, monitors should be independent of what they monitor, and the right to reproduce should sit outside the thing that would want to use it. Alignment is cancer suppression for cyberphysical collectives.