RegLegBrief › About

About RegLegBrief

RegLegBrief is the published, primary-source-verified delta between AI-generated accounts of regulation and the regulator's own text on the regulator's own portal. Last updated: 2026-06-06.

Bottom line. RegLegBrief retrieves each regulator's verbatim text from the regulator's own portal, tests named AI subjects against the same rule, classifies every disagreement by failure mode, and publishes the result with an immutable Citation ID. Operated by Verdus Technologies Pte. Ltd., Singapore.

What we do, in one paragraph

AI models reliably mis-state regulatory rules in ways the user cannot detect from the output alone. The regulator's primary source is the only ground truth that resolves the disagreement. RegLegBrief retrieves each regulator's verbatim text from the regulator's own portal, tests named AI subjects against the same regulation through an asymmetric question battery, classifies every disagreement by failure mode, and publishes the result with an immutable Citation ID. The catalogue covers 21 regulations today and grows on a continuous cadence.

Methodology, in five steps

# Step What happens
1 Substrate construction The regulator's verbatim text is retrieved from the regulator's own portal (MAS, FCA, BIS/CPMI, CFTC, HKMA, IMF, …) and archived against tampering. No secondary sources, no aggregator feeds.
2 Asymmetric question design A structured question battery is built per regulation — covering the rule's deontic structure, named entities, scope carve-outs, threshold values, and the specific traps AI models tend to fall into.
3 Multi-Subject audit Named AI subjects (frontier model + tool variant, e.g. Opus-4.7 with web search, Sonnet-4.6 with web search) answer each question. Outputs are captured verbatim.
4 Specialist Panel verification Each AI answer is compared against the substrate. Deltas are classified by failure mode (deontic-register inversion, negation-reversal, schema substitution, entity misidentification, …) and rated for materiality.
5 Publication with Citation ID Material findings are published with an immutable Citation ID (RLB-F-… or RLB-H-…) that anchors the finding to the verifiable record — source, AI subject, question, delta, failure mode.

Read the full methodology for the rules that govern what can and cannot be published.

What gets published

Primary sources we verify against

The regulator's own portal is the only ground truth. Every finding cites verbatim text from one of these primary sources:

The Specialist Panel

Every published finding is verified by the RLB Specialist Panel before it appears on the site. The Panel is independent of the AI subjects under test and independent of the regulators whose rules are tested. The Panel applies the same standards a senior practitioner would apply to a court submission — verbatim source citation, exact quotation, named attribution. No finding is published without primary-source substrate behind it; this is the published no-substrate-no-audit rule.

Full Panel composition, domain coverage, codified verification rules, classification taxonomy, and independence statement are documented on the Specialist Panel page.

Who reads this

Four audiences depend on getting regulatory questions right and use RLB findings differently:

Owned and operated by Verdus Technologies Pte. Ltd.

RegLegBrief is owned and operated by Verdus Technologies Pte. Ltd., a private technology company incorporated in Singapore (UEN: 201616982R). Verdus Technologies builds and operates platforms in the regulatory intelligence space.

For platform enquiries: [email protected] · Full contact list

Important. RegLegBrief publishes AI hallucination research verified against authenticated regulatory primary sources. Each finding is the documented delta between an AI subject's account of a rule and the regulator's verbatim text, with the failure mode classified and the Citation ID published. The methodology, the primary source, and the verification trail are all linked from every finding. Reading both the finding and the cited primary source is the audit structure the site is built around.

Common questions

What does RegLegBrief do?

Tests named AI subjects against authenticated regulatory primary sources, classifies failures by mode, and publishes findings with immutable Citation IDs. The regulator's own portal is the only ground truth.

Who reads RegLegBrief?

AI Labs (remediation input), Licensed Practitioners (independent verification before relying on AI output), Regulated Firms including banks (integration into AI governance and MRM), and Regulators (right of reply on findings affecting their rules).

Who owns and operates RegLegBrief?

Verdus Technologies Pte. Ltd., a private technology company incorporated in Singapore (UEN 201616982R). RegLegBrief is its primary product.

How is each finding verified?

By the RLB Specialist Panel against the regulator's verbatim text from the regulator's own portal. The no-substrate-no-audit rule applies — no finding is published without primary-source substrate behind it.

← Back to RegLegBrief