Communication Bias in Large Language Models: A Regulatory Perspective