Write-Before-Execute: A Governance Standard for Evidentially Defensible AI Decision Logging in Regulated Enterprises

When an autonomous AI system causes material harm and a regulator asks for evidence of what the system decided and why, most organisations discover a structural problem in their logging architecture: the log they produce was created after the action was executed. It describes what happened. It does not constitute evidence that a governance state existed before the consequence. This distinction between a record and evidence is the central governance problem that the Write-Before-Execute standard addresses. This paper introduces Write-Before-Execute as a governance standard that requires every consequential autonomous AI decision to generate a complete log entry, including a cryptographic hash of the log content, which is committed to immutable, append-only storage before the action executes. If logging infrastructure is unavailable, the system does not proceed. The log is a condition of execution, not its record. This architectural requirement is not merely a best practice for operational integrity. It is the structural precondition for a regulated enterprise to demonstrate, under regulatory examination, that its governance controls were operative at the time of each decision. The paper establishes the evidentiary basis for the Write-Before-Execute requirement, drawing on evidence, law, operational risk governance, and the systems engineering literature on safety-critical logging. It distinguishes Write-Before-Execute from post-execution logging architectures across four dimensions: evidential defensibility, tamper resistance, governance state capture, and regulatory examination readiness. It introduces the Write-Before-Execute Compliance Standard, specifying the mandatory log fields, the cryptographic commitment mechanism, the infrastructure availability obligation, and the minimum compliance threshold. It demonstrates that the major existing governance frameworks — ISO/IEC 42001:2023, the EU AI Act, APRA CPS 230, the NIST AI RMF, and the IMDA Model AI Governance Framework for Agentic AI — require logging but do not require pre-execution commitment. This paper argues that the gap is not a drafting oversight but a structural omission with material regulatory consequences, and that Write-Before-Execute closes it.