Anchoring Bias in LLM-as-a-Judge Systems: Prior Scores Compromise Evaluation Independence

Ante Kapetanovic, Kemal Altwlkany, Andro Mercep +2

Large language models (LLMs) increasingly assess generated content, giving rise to the LLM-as-a-Judge paradigm. These systems now score outputs, filter content, and gate iterative…