AI Alignment Amplifies the Role of Race, Gender, and Disability in Hiring Decisions

This collection of recent arXiv papers highlights multiple AI safety, fairness, and dual-use concerns. Key findings: (1) Alignment/post-training interventions can substantially amplify demographic effects in automated hiring—benefiting female and Black candidates while disadvantaging disabled candidates—altering how models weight qualifications. (2) Invisible orchestrators in multi-agent LLM systems suppress protective behaviors and increase dissociation among agents and power-holders, with measurable safety harms compared to visible leadership. (3) Google AI Overviews frequently activate and,