When Can an AI Answer?

When Should an AI Be Allowed to Answer? Most large language models are evaluated by how well they respond. This work focuses on a quieter but more consequential question: when should an AI respond at all? We present a verifiable governance protocol that grants an AI system the Right to Answer only when a minimal evidence chain satisfies explicit policy constraints. Model outputs are treated as untrusted proposals; answer delivery is a privileged operation enforced by a minimal governance kernel. At the core of the protocol is Reverse Evidence Chain Inference (RECI), which frames governance not as a matter of confidence scores, but as a feasibility problem: determining whether there exists a smallest possible set of evidence that meets coverage thresholds, consistency constraints, and risk budgets. If such a set exists, delivery is authorized; if not, the system is required to ask or abstain. The protocol further specifies an audit interface, including environment fingerprints, structured reason codes, and deterministic replay commands, enabling post hoc verification of every delivery decision. No new learning algorithm is proposed; the contribution lies at the protocol and system boundary, where accountability is enforced not by trust, but by replayable constraints. This record targets high-stakes AI deployments—such as medicine, law, and public services—where answering the wrong question confidently is often worse than not answering at all.