Safety consultations for AI lab employees — LessWrong