The platform For Canada’s federally regulated financial institutions
OSFI software for models, AI and vendors, on one record.
MINERVA is governance infrastructure for FRFIs: one record for what OSFI’s E‑23, B‑10 and B‑13 ask about your models, AI systems, AI agents and suppliers, with decisions enforced and evidence your examiner can verify without trusting us.
01 The guidelines, and what the platform does
Each guideline is explained in plain words: E‑23, B‑10 for AI vendors, B‑13, the AI bulletin and the AMF’s guidelines.
| The guideline | What MINERVA does |
|---|---|
| E‑23, Model Risk Management (in effect 1 May 2027) | The model inventory in E‑23's fields, a risk rating that sets each model's cadence, independent review, approvals held until the evidence is on file, monitoring against thresholds, and the evidence packet. |
| B‑10, Third-Party Risk Management (in effect 1 May 2024) | Every supplier, AI vendors included, by risk and criticality; due diligence to exit plan; the models and agents that depend on each; vendor incidents strung to every dependant. |
| B‑13, Technology and Cyber Risk Management (in effect 1 January 2024) | The AI and model part of it: an inventory of AI systems and agents with owners and dependencies, changes decided with duties separated, AI incidents routed and escalated. Not security operations. |
| The July 2026 bulletin on generative and agentic AI | Each agent its own identity, a least-privilege policy checked before every tool call, high-impact calls held for a person, a kill switch, and every decision logged. |
| Quebec's AMF: Model Risk Management (in effect since June 2025) and the Use of AI (in effect 1 May 2027) | The same record: AI systems in the inventory with a risk-based rating and a decided lifecycle, in English and French. No AMF-specific rule pack yet. |
02 One record for every guideline
- One inventory
- Models, AI systems, AI agents, datasets and suppliers on one list, each with an owner and what depends on it.
- Governance
- Board and senior management's approval authority, three lines of defence, and the roles each guideline names, recorded and enforced at the moment of decision.
- Risk appetite
- The appetite and tolerance recorded; risks over tolerance flagged; acceptance decided by someone other than the person asking.
- Decisions
- Nothing changes itself: a change is proposed, a named person decides, and a developer never approves their own work.
- Evidence
- Every change attributed, hash-chained and time-stamped by two independent authorities; one packet the examiner verifies offline; a read-only examiner login.
- Where it runs
- MINERVA's cloud, hosted in Canada, or your own data centre. English and French.
03 The platform, by job
OSFI E‑23 platform for model risk · Agentic AI governance platform for AI agents · AI vendor risk software under B‑10 · E‑23 platforms compared, ours beside SAS, IBM, ValidMind, Yields and others.
04 What it does not do
MINERVA does not make an institution compliant or issue a certification. It is not a cyber security tool, a core banking system or a model development environment, and it does not host model validation compute. Questions to ask us and every other vendor are in our buyer’s guide.
05 Questions people ask
What is OSFI software?
Software a federally regulated financial institution uses to run and evidence what OSFI's guidelines expect, such as model risk management under E-23, third-party risk under B-10, and technology and cyber risk under B-13. OSFI does not prescribe a tool; most institutions use several.
Does OSFI approve or certify software?
No. OSFI does not approve, endorse or certify software. MINERVA issues no certification either: it keeps the record and the evidence, and the institution and its examiners judge it.
Which OSFI guidelines cover AI?
E-23 covers AI and machine-learning models from 1 May 2027. B-10 covers AI vendors as third parties. B-13 covers the technology AI runs on. OSFI's July 2026 technology risk bulletin sets out sound practices for generative and agentic AI that build on B-13, E-21 and B-10.
Can one platform cover E-23, B-10 and B-13?
For the record they share, yes: one inventory of models, AI systems, agents and suppliers, one set of owners and decisions, one trail. B-13's security operations, such as vulnerability scanning, patching and network monitoring, belong to security tools, and MINERVA does not do them.
Does MINERVA cover Quebec's AMF guidelines?
The record covers the same ground the AMF's model risk and AI guidelines ask about: an inventory of AI systems, a risk-based rating and a governed lifecycle, in French and English. MINERVA ships rule packs for OSFI E-23 and AI risk, not yet one written from the AMF's text.
Is it only for banks?
No. It is for every FRFI, banks, insurers and trust and loan companies, and for credit unions and provincially regulated institutions whose regulators ask for the same model and AI governance.
What does it cost?
A 30-day pilot on your real inventory is CAD 15,000, credited to year one. A subscription is CAD 36,000 to 150,000 a year by register size, with unlimited seats.
Also: OSFI E‑23 platform · OSFI B‑13, explained · The AMF’s AI and model risk guidelines · Know Your Agent · Policy as code.