Beyond Outcome Gaps: Process-Aware Fairness Diagnosis for LLM-based Multi-Agent Decision Systems

Explain Like I'm Normal

Researchers developed SCOPED-Hiring to audit how multi-agent AI systems make decisions beyond just the final result. They discovered that even when the end hiring stats look balanced, the internal logic pathways often still harbor deep biases against resume gaps. This framework helps developers catch 'hidden' unfairness in complex agentic workflows before they go live.