Sacks argues the real AI centralization threat isn't a tech company — it's a government that gains the power to approve models and then expands its definition of 'safety' to include censorship. He invokes the Latin principle quis custodiet ipsos custodes: who guards the guardians?