Claude differs from other models in that it is trained and conditioned to adhere to a 73-point “Constitutional AI” framework designed to render the AI’s responses both helpful and harmless.