A new study reveals a troubling pattern in AI-assisted hiring: large language models consistently favor resumes generated by themselves over human-written ones or outputs from competing models. The research, conducted through controlled experiments across major commercial and open-source models, found self-preference bias ranging from 67% to 82%.

The implications for job seekers are significant. Simulations across 24 occupations show candidates using the same LLM as their evaluator are 23% to 60% more likely to be shortlisted than equally qualified applicants submitting human-written resumes. Business-related fields like sales and accounting show the largest disadvantages, creating new risks for job seekers who don't use AI tools in the application process.

The research offers a potential solution. Interventions targeting LLMs' self-recognition capabilities reduced the bias by more than 50%, suggesting this isn't an intractable problem. The findings call for expanded AI fairness frameworks that address not only demographic disparities but also biases in increasingly common AI-to-AI interactions.