Model Agentic AI Evidence and Black-Box Accountability Law Drafted by DR Pavan Duggal

This draft Model Law proposes a legislative and regulatory reference text on evidence in agentic and autonomous artificial intelligence. It addresses what records an agentic AI system should generate, who should preserve them and for how long, how they may be proved in court, and what a court or authority may infer when evidence that ought to exist is missing. Its premise is that a black box may limit explanation but must not eliminate accountability. It does not require disclosure of model weights or reconstruction of every computational step. It asks instead for legally sufficient accountability evidence: what the system did, under whose authority, with what information and tools, subject to what human intervention, and with what result. The text has 90 sections in 40 Parts. It covers the agentic event record and chain of evidence, memory and state evidence, black-box accountability and the Black-Box Accountability Doctrine, admissibility, chain of custody, duty to preserve, tiered disclosure, trade secret and privacy safeguards, multi-agent, tool and API evidence, incident forensics, provider and deployer duties, certification, cross-border cooperation, and the rights of affected persons. Annex A is an illustrative implementation matrix and Annex B sets out four hypothetical applications. The Model Law is offered as the evidentiary companion to the Duggal Global Agentic AI Liability Framework. It is a draft reference instrument offered for comment. It is not binding law in any jurisdiction, and the views expressed are the author’s own.