New Framework for Epistemic Trustworthiness in Generative AI

A recent study published on arXiv (2608.05602) introduces a normative framework intended to assess when it is justified to depend on generative AI outputs, especially in critical professional settings. The authors contend that while metrics such as accuracy, fairness, explainability, safety, and user trust are essential, they fall short in clarifying the circumstances under which users can rightfully consider AI outputs as part of their reasoning process. By referencing philosophical perspectives on trustworthiness, defined as competence and audience focus, the paper articulates epistemic trustworthiness as the characteristic that renders a system worthy of reliance. This framework seeks to promote responsible AI development and use by outlining specific evaluative criteria for justified reliance. The paper is listed as an abstract announcement on arXiv under the identifier 2608.05602v1.