A University of Queensland study has shown large language models (LLMs) used in AI content moderation may be prone to subtle biases that undermine their neutrality. A team led by data scientist ...