Architecture Standard · Not a Feature

Regulatory-Native
AI

Most "AI-native compliance" tools apply AI to compliance workflows while the decision underneath stays a black box, with explainability bolted on after the fact. Sidian inverts that: determinism, audit trails, and explainability are structural properties of the decision itself — not features added to satisfy a reviewer.

The Gap

AI-Native Compliance vs. Regulatory-Native AI

The distinction is architectural, not semantic.

Property
AI-Native Compliance
Regulatory-Native AI
Core decision logic
General-purpose ML, often a black box to its own vendor
Deterministic where possible; explainable-by-construction elsewhere
Explainability
Post-hoc wrapper bolted onto a finished model
Co-developed with the model, from day one
Audit trail
Application log, queried after the fact
Write-ahead log and immutable manifest, first-class outputs
Regulator relationship
Regulator audits the bank's use of the tool
Vendor is in supervisory scope directly, and designs for it
The Standard

Five Criteria for Regulatory-Native AI

Guardian SGSV already satisfies the first four. The fifth — the explainability layer for RegOS's ML components — is an active, phased build.

Deterministic-First
Where a decision can be an explicit, auditable rule, it is — not learned implicitly by a model.
Fail-Closed by Default
Uncertainty routes to a block or human escalation, never to a silent pass. A design default, not an exception handler.
Audit-Native, Not Audit-Added
The audit trail is a direct output of the decision process itself, not reconstructed from logs after the fact.
Human Override Is Structural
Escalation and quarantine are load-bearing, governed by quorum controls — not a button that can be silently bypassed.
Explainability Co-Developed
For ML components, feature attribution and plain-language rationale are built in parallel with the model — never retrofitted.
What's Live Today

The standard isn't just a thesis — here's what's actually built.

Guardian SGSV
Live & CI-tested
Deterministic gate engine running today — not a whitepaper architecture.
IP Position
Patent-pending
IPOS application filed with Singapore's Intellectual Property Office.
MAS Engagement
Active
Ongoing dialogue with MAS's FinTech Office on this standard.
RegOS Explainability
Funded, active build
The fifth criterion above — in progress toward MAS FEAT-defensible production status.
Regulatory Timing

This Is Not a Speculative Trend

A concrete, near-term forcing function — for financial institutions and for the vendors supplying their AI systems.

Nov 2025 – Jan 2026
MAS Consultation P017-2025
Proposed AI Risk Management Guidelines for Financial Institutions — closed 31 January 2026, bringing technology vendors supplying AIDA solutions into supervisory scope, not just the banks deploying them.
Jan 2026
IMDA Model AI Governance Framework for Agentic AI
Singapore's first named regulatory framework specifically for autonomous AI agents, released 22 January 2026 — the direct national-level counterpart to MAS's P017-2025 consultation, and the exact class of system Guardian is built to govern at the point of execution.
Mar 2026
Project MindForge Phase 2
Operationalisation Handbook extends MAS's FEAT principles and Veritas toolkit to generative and agentic AI.
Jul 2026
MAS SAFR — Safeguards for Agentic Finance at Runtime
MAS's BuildFin.ai programme, co-authored with HSBC, J.P. Morgan Chase, Mastercard, OCBC and other industry partners, publishes the reference architecture for governing AI agents at runtime — identity, controls, disposition, and audit.
Aug 2026
EU AI Act — High-Risk Provisions
High-risk provisions, including credit-decisioning use cases, apply from 2 August 2026 — reinforcing explainability as a cross-jurisdictional requirement.
FAQ

Questions About the Standard

Is this a real regulatory requirement, or Sidian's own framing?

Both MAS Consultation P017-2025 and IMDA's Model AI Governance Framework for Agentic AI are live, dated regulatory activity — not projections. Regulatory-Native AI describes the direction that activity is converging toward, ahead of final codification, not a claim that any regulator has adopted Sidian's specific terminology.

How does this relate to MAS's SAFR framework?
How is this different from other “AI compliance” vendors?
Is Sidian's architecture protected?
Where is this used today?
First Cohort — Forming Now
Early is the advantage, not the risk.
Founding Partners help shape the standard and lock in commercial terms before Series A — the same first-mover position Guardian SGSV gives you with MAS.
Apply for Founding Partner Access →

Building on this architecture

Guardian SGSV, Sentry, and RegOS implement this standard today. If you're a MAS-regulated institution evaluating AI decisioning infrastructure, we'd like to talk — or start with an independent read on your existing system.