“ML models must also think about trusting trust” (lightbluetouchpaper.org)