AI Chatbots’ Hidden Scoring Systems Could Affect Users’ Credit and Job Prospects

Large language models are already used for résumé screening, credit assessments and medical advice, potentially forming a type of “effective trust” in decision-making contexts. Research from the Hebrew University of Jerusalem found that models score competence, integrity and benevolence separately. If characteristics such as age and gender seep into those judgments, they could directly affect credit limits and employment opportunities.

The study was published in Proceedings of the Royal Society A in 2026, released by the university on April 13 and reported by TechNews on April 17. The team compared five LLMs across five scenarios, conducting 43,200 simulations and involving about 1,000 participants. Ages 20, 40 and 60, as well as gender and religion, altered loan and donation recommendations, though the study did not identify a single figure for the monetary difference.