Research discovered that generative AI during training vouches for AI alignment but then once fielded violates those precepts ...