A new study highlights the growing risks of bias in large language models (LLMs) as they become cheaper and widely used, calling for ethical AI policies to ensure fairness and transparency.