Controls
The reference policy evaluates 24 controls across B-13, B-10, E-21 and E-23. This page is generated from that policy — the control catalogue, the Rego rules and mapping.yaml — so it cannot describe something the gate does not do.
osfi-gate has no opinion about your controls. It evaluates a policy your second line owns and versions in Git. What follows is the policy this repository ships with. Change it and this changes with it, which is the point.
Blocks, or only warns
The distinction a coverage table usually loses. A control that warns is not a control that blocks, and rendering both as a tick is the kind of thing this project exists to stop.
- 24 block. A finding fails the run, and the change does not merge without a waiver that risk approved in Git.
8 do both, and the split is where the policy does its real work — each blocks on one condition and warns on an adjacent one:
- CM-01 — Changes are approved by the required number of reviewers: 1 blocking, 2 warning.
- MR-02 — AI and model usage is registered in the model inventory: 3 blocking, 2 warning.
- MR-04 — Models are pinned to the validated version: 2 blocking, 1 warning.
- MR-06 — Governance controls match the model's materiality: 2 blocking, 1 warning.
- MR-07 — Independent validation is current: 3 blocking, 1 warning.
- MR-08 — Model changes are approved by the second line: 4 blocking, 1 warning.
- WV-01 — Risk acceptances are approved, bounded and current: 4 blocking, 2 warning.
- EV-01 — Evidence collection is complete: 7 blocking, 1 warning.
MR-02 is the one worth reading closely. AI usage the change introduces blocks; usage that was already there warns. That is what makes it possible to enrol a repository with an existing estate without stopping every build on day one — and the pre-existing usage stays visible instead of being waived away.
The thresholds
Policy values, not product behaviour. These are what the reference policy ships with:
| Setting | Value |
|---|---|
| Independent approvals required | 2 |
| Severities that block | critical, high |
| Vulnerability SLA, in days | critical 7 · high 30 · medium 90 · low 180 |
| Maximum waiver term | 180 days |
| Who may approve a waiver | risk-officer, risk-lead |
| Denied licences | GPL-*, AGPL-*, LGPL-*, SSPL-*, EUPL-*, OSL-*, CC-BY-NC-* |
Change management and segregation of duties
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| CM-01 Changes are approved by the required number of reviewers |
B-13 Domain 2, 2.5 Change and release management | blocks | PR approvals on the head commit >= gate.min_approvals |
| CM-02 Authors cannot approve their own change |
B-13 Domain 2, 2.5 Change and release management | blocks | PR author not in approvers |
| CM-03 Target branch is protected |
B-13 Domain 2, 2.5 Change and release management | blocks | Branch protection enabled with required approvals >= gate.min_approvals |
| CM-04 Commits are signed and verified |
B-13 Domain 3, 3.2 Defend | blocks | Every PR commit has verification.verified = true |
| CM-05 Changes are traceable to a reviewed pull request |
B-13 Domain 2, 2.5 Change and release management | blocks | Push events resolve to a merged PR |
Secure development
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| SD-01 Static analysis has no unwaived high or critical findings |
B-13 Domain 2, 2.4 System development life cycle | blocks | Semgrep findings with severity in gate.blocking_severities Fed by: semgrep |
Vulnerability management
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| VM-01 Known vulnerabilities are remediated within SLA |
B-13 Domain 2, 2.6 Patch management; Domain 3, 3.1 Identify | blocks | Trivy vulnerability age <= gate.sla_days[severity]; the default branch is rescanned daily Fed by: trivy |
Technology asset management
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| AM-01 A software bill of materials is produced for every build |
B-13 Domain 2, 2.2 Technology asset management | blocks | SBOM present, parseable, non-empty, digest recorded |
Secrets and cryptography
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| SC-01 No secrets are committed to source |
B-13 Domain 3, 3.2 Defend | blocks | Gitleaks (current tree and the change's commit history), Checkov secrets and Semgrep secret rules report nothing unwaived Fed by: checkov, gitleaks, semgrep |
| SC-02 Only approved cryptographic algorithms are used |
B-13 Domain 3, 3.2 Defend | blocks | osfi-sc02- Semgrep rules report nothing unwaived Fed by: semgrep* |
Configuration baseline
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| CB-01 Infrastructure meets the security configuration baseline |
B-13 Domain 2, 2.1 Technology architecture; Domain 3, 3.2 Defend | blocks | Baseline Checkov/Trivy/Semgrep IaC findings with severity in gate.blocking_severities Fed by: checkov, semgrep, trivy |
Third-party components
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| TP-01 Dependency licences comply with policy |
B-10 Third-Party Risk Management | blocks | SBOM component licences not matching gate.denied_licenses Fed by: osfi-gate |
| TP-02 Dependencies are pinned and sourced from approved registries |
B-10 Third-Party Risk Management; B-13 Domain 2, 2.4 | blocks | Lockfiles present, requirements pinned, FROM/uses pinned, resolved hosts approved Fed by: osfi-gate, semgrep |
Operational resilience
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| OR-01 Critical services declare resilience configuration |
E-21 Operational Resilience; B-13 Domain 2, 2.9 Disaster recovery and testing | blocks | osfi-or01- Semgrep rules and Checkov CKV_AWS_133/144/157 Fed by: checkov, semgrep* |
Model governance
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| MR-01 Model artifact changes reference inventory and validation |
E-23 Model Risk Management | blocks | Changed model files listed in model-inventory.yaml with inventory_id and validation_record |
| MR-02 AI and model usage is registered in the model inventory |
E-23 Model Risk Management; B-10 Third-Party Risk Management | blocks | osfi-mr02- detections joined to model-inventory.yaml; a change may not introduce unregistered usage Fed by: semgrep* |
| MR-03 AI inference stays with approved providers and regions |
B-10 Third-Party Risk Management; B-13 Domain 1, Governance and risk management | blocks | Provider hosting class in gate.ai.approved_hosting and declared region in the provider's region list |
| MR-04 Models are pinned to the validated version |
E-23 Model Risk Management | blocks | Model literal not in gate.ai.floating_model_patterns and equal to the inventory entry's model |
| MR-05 Every model has a current materiality assessment |
E-23 Model Risk Management, 3.2 Model risk rating | blocks | All gate.ai.e23.dimensions rated, assessed_by present, assessed_on within gate.ai.e23.assessment_max_age_days |
| MR-06 Governance controls match the model's materiality |
E-23 Model Risk Management, 5 Model lifecycle; HGAI Principle 4 (human oversight) | blocks | gate.ai.e23.required_controls for the computed tier, plus genai_required_controls for generative model types, each recorded as "yes" |
| MR-07 Independent validation is current |
E-23 Model Risk Management, 5.3 Model validation | blocks | validation.record and validation.validated_on present, validation.next_review in the future |
| MR-08 Model changes are approved by the second line |
E-23 Model Risk Management, 4 Governance and oversight; B-13 Domain 1, Governance and risk management | blocks | metadata.model_risk records an approved assessment for each introduced call site; unreachable or uncaptured is not approval |
Gate governance
| Control | Anchored to | Behaviour | What it checks |
|---|---|---|---|
| WV-01 Risk acceptances are approved, bounded and current |
B-13 Domain 1, Governance and risk management | blocks | Waiver schema valid, not expired, term <= gate.max_waiver_days, waiver changes approved by gate.risk_approvers |
| EV-01 Evidence collection is complete |
B-13 Domain 1, Governance and risk management | blocks | All gate.required_scanners completed (skips only on manual runs), platform metadata collected, vulnerability DB metadata recorded and newer than gate.max_vuln_db_age_days |
What this page is not
It is a description of the reference policy, not of your institution's. An organisation running its own policy repository will have different thresholds, different waiver approvers and possibly different controls, and that is the intended way to use this.
It also says nothing about whether a control is well implemented for you. The gate records what the policy decided; whether the policy was right is a question about the policy, which is why it is versioned, reviewed and identified by commit in every bundle. See the audit trail.