What it does

Canadian securities regulators and FINRA have both made their expectations around AI more explicit.

CSA Staff Notice and Consultation 11-348, published 5 December 2024, explains how existing securities law applies when firms use AI systems. It creates no new legal requirements. It sets out that registrants must maintain records demonstrating compliance, and that AI systems should provide enough explainability for a firm to meet its record-keeping obligations. It names low-explainability systems as challenging transparency, accountability, record keeping and auditability.

FINRA's 2026 Annual Regulatory Oversight Report, published December 2025, raises agentic AI for the first time. It is examination-staff observation rather than rulemaking. The supervisory questions it puts to US broker-dealers are recognisable: agent autonomy and scope of authority, auditability, where human oversight belongs, tracking what an agent did, and guardrails on what a system may do.

Neither creates a new duty. Both make the firm responsible for the output of the technology it uses.

Most approaches to this produce a document: a policy describing how the model is meant to behave, a validation report describing how it behaved in testing, an explanation generated on request after someone asks.

The problem with a reconstruction is that it cannot establish what actually shaped the original decision. Information that arrived later, or a system state that has since changed, contaminates it. It defeats a reviewer for the same reason it defeats an examiner.

Orbiqent produces the record as a byproduct of the system operating. The reasoning exists because the decision was made, not because someone asked afterwards.