Cross-Examining The Code: The Necessary Art of the AI Voir Dire

Abstract

The legal profession is shifting from deterministic legal research tools to probabilistic generative systems capable of producing fluent but false legal outputs. This article proposes AI Voir Dire as a structured reliability protocol for the legal use of artificial intelligence. The concept is a procedural metaphor, not a claim that AI is a legal witness. It adapts the logic of voir dire – testing competence, bias, credibility, and admissibility – to AI-assisted legal work and AI-influenced evidence. The article distinguishes six legally distinct uses of AI: legal research and drafting, expert analysis, machine-generated evidence, synthetic media, algorithmic risk assessment, and litigation analytics. It then develops four inquiries: professional competence, methodological reliability, authenticity and provenance, and bias and due process. Recent U.S. cases show that courts are not rejecting AI categorically, but are rejecting uses that are unverified, unexplained, undisclosed, or unfairly prejudicial. The article argues that responsible AI integration requires task-specific validation, preserved audit trails, independent citation verification, disclosure where required, and nondelegable human accountability. It concludes with an AI Voir Dire protocol for legal filings, expert evidence, synthetic media authentication, and courtroom admissibility.