Anthropic has produced it as clear as possible that it'll under no circumstances use a user's prompts to coach its versions Until the person's conversation has actually been flagged for Trust & Security overview, explicitly claimed the supplies, or explicitly opted into instruction.