Courtrooms are paradigmatic formal epistemic environments: they are structured by evidentiary rules, institutional roles, and procedural norms designed to protect the integrity of fact-finding. Yet courts are also social institutions and thus vulnerable to the same malicious epistemic interventions that destabilize less formal social epistemic environments. As generative media systems proliferate, courts will increasingly confront not only fabricated audiovisual artifacts (“deepfakes”), but also the broader erosion of epistemic trust in digital media that has long functioned as a gold-standard evidentiary artifact. The problem is therefore not merely misinformation, but the loosening of epistemic norms as trust breaks down (Rini, 2020).

Legal scholarship has responded by emphasizing (i) provenance and forensic practices within existing doctrine, (ii) expanded judicial gatekeeping over authenticity, and (iii) stronger sanctions and professional-responsibility constraints on opportunistic “deepfake defenses” (Pfefferkorn, 2020; Delfino, 2023, 2024; Dalal et al., 2024). This Article argues that these proposals must be integrated within a social-epistemic approach to adjudication. In addition to provenance-based indicators (chain of custody, metadata integrity, capture and transfer history), judges should adopt structured, procedurally constrained assessments of agent-centered reliability: the credibility and incentives of sponsoring parties and counsel, their diligence in preserving provenance, consistency with the broader record, and litigation conduct. A two-track protocol—artifact-centered and agent-centered—aims to reduce both false admissions of fakes and strategic discounting of authentic evidence while preserving adjudicative legitimacy under deepfake uncertainty.