Skip to content

Coverage matrix

How each Regulus mechanism maps to each shipped profile, with the ADK hook it uses. Regenerate with ./gradlew regulusComplianceMatrix.

How to read this

  • Mechanism — the Regulus control name (matches ControlBinding.mechanism).
  • Profile clauses — which articles / paragraphs each profile cites the mechanism against.
  • ADK hook — the official ADK extension point Regulus implements it through.
  • Test fixture — where the unit test for this mechanism lives.

Matrix

Mechanism EU AI Act GDPR / UK GDPR DORA NIS2 FCA SYSC PRA SS1/23 PRA SS2/21 NHS DSPT EHDS ADK hook Test fixture
purpose-binding — Art. 5(1)(b) — — — — — — — BeforeModelCallback (PolicyPlugin) DefaultPolicyEngineTest#purposeBinding
automated-decisions-safeguards — Art. 22 — — — — — — — BeforeModelCallback (PolicyPlugin) + ToolConfirmation DefaultPolicyEngineTest#art22Safeguard
privacy-by-design — Art. 25 — — — — — — — BeforeModelCallback (PrivacyPlugin) RegulusPrivacyPluginTest#redact
pii-redaction — Art. 5(1)© — — — — — DSPT 1.x — BeforeModelCallback + AfterModelCallback (PrivacyPlugin) BuiltInPatternsTest
storage-limitation Art. 19 Art. 5(1)(e) Art. 12 — SYSC 9 — — NHS records EHDS retention EventCompactor (RetentionEventCompactor) RegulusRetentionEventCompactorTest
audit-trail Art. 12 Art. 30 Art. 12 Art. 23 SYSC 9 SS1/23 §2-5 SS2/21 §7 DSPT 6.x Art. 56 After*Callback (AuditPlugin) RegulusAuditPluginTest
cross-border-residency — Arts. 44-49 Art. 28 — SYSC 13 — SS2/21 §6 DSPT 8.x Chapter III Startup + BeforeAgentCallback (DataResidencyPlugin) RegulusDataResidencyPluginTest
dual-control-kill-switch Art. 14 — — — (FG22/5) SS1/23 §6 — — — BeforeAgentCallback (KillSwitchPlugin) + ToolConfirmation InMemoryKillSwitchStoreTest
model-risk-tier Art. 9, Annex III — — — SYSC 4 SS1/23 §3 — — — BeforeModelCallback, BeforeToolCallback (ModelRiskPlugin) RegulusModelRiskPluginTest
transparency-disclosure Art. 13 — — — — — — — — AfterModelCallback (AuditPlugin) RegulusAuditPluginTest#transparency
post-market-monitoring Art. 16 — — — — SS1/23 §5 — — — AfterAgentCallback (AuditPlugin) + observability (integration)
deployer-obligations Art. 26 — — — — — — — — BeforeAgentCallback (KillSwitchPlugin + PolicyPlugin) (integration)
incident-classification — Art. 33 Arts. 17-23 Art. 23 — — — DSPT 6.x — AfterAgentCallback (AuditPlugin) RegulusAuditPluginTest#incidentSeverity
third-party-risk — — Arts. 28-30 Art. 21(2)(d) SYSC 13 — SS2/21 §3 — — Model registry + audit linkage (integration)
consumer-duty-good-outcomes — — — — FG22/5 — — — — BeforeModelCallback (PolicyPlugin) DefaultPolicyEngineTest#consumerDuty
vulnerable-customer-handling — — — — FG22/5 §4 — — — — BeforeModelCallback (PolicyPlugin) + ToolConfirmation DefaultPolicyEngineTest#vulnerableCustomer
senior-management-arrangements — — — — SYSC 4 — — — — Audit attribution (smf_holder) RegulusAuditPluginTest#smfAttribution
model-inventory — — — — — SS1/23 §2 — — — Model registry ModelRegistryTest
kill-switch-readiness Art. 14 — — — — SS1/23 §6 — — — BeforeAgentCallback (KillSwitchPlugin) RegulusKillSwitchPluginTest
personal-data-protection — — — — — — — DSPT 1.x — BeforeModelCallback (PrivacyPlugin) BuiltInPatternsTest#nhsNumber
clinician-identity — — — — — — — DSPT 4.x — Audit attribution (clinician_smartcard_id) RegulusAuditPluginTest#clinicianIdentity
secondary-use-permit — — — — — — — — Chapter IV Audit attribution (permit_ref) (integration)
data-quality-labels — — — — — — — — Art. 56 AfterModelCallback (AuditPlugin) (integration)

Framework bindings

The same Regulus mechanisms above also bind to governance framework control ids. Activate via regulus.governance.frameworks: [...].

Mechanism NIST AI RMF NIST 600-1 GenAI NIST Agent Interop (planned) ISO/IEC 42001 ISO/IEC 23894 ISO/IEC 23053
purpose-binding MAP-1.1 — — A.9.2 — —
automated-decisions-safeguards — GAI-7 — — — —
privacy-by-design — GAI-4 — A.7 — —
pii-redaction — GAI-4 — A.7.3 — —
storage-limitation GOVERN-1.5 — — A.6.2.7 — —
audit-trail GOVERN-1.5, MEASURE-1.1 GAI-8 AGENT-MONITORING-1 A.6.2.7 CL-6.7 CL-6.5
cross-border-residency MEASURE-2.7 GAI-9 — A.6.2.4 — —
dual-control-kill-switch GOVERN-4.1 GAI-2 AGENT-SECURITY-2 A.6.2.8 CL-6.6 —
model-risk-tier MAP-4.1 GAI-7 AGENT-SECURITY-1 A.5.2 CL-6.3, CL-6.4 CL-7
transparency-disclosure MEASURE-2.8 GAI-7 — A.8.5 — —
post-market-monitoring MANAGE-4.1 — — A.6.2.5 — —
incident-classification MANAGE-2.2 — — A.8.4 — —
third-party-risk GOVERN-6.1 GAI-12 — A.10.3 — —
senior-management-arrangements GOVERN-2.1 — — A.3.2 — —
model-inventory — — — A.4.4 — CL-6.2
policy-engine GOVERN-1.1 GAI-3 AGENT-IDENTITY-2 A.2.2 — —
a2a-envelope — — AGENT-IDENTITY-1, AGENT-MONITORING-2 — — —

Bindings are generated from each GovernanceFramework.bindings(). The NIST AI RMF Agent Interop Profile column uses provisional IDs from the April 2026 concept note; final IDs land when NIST publishes (target Q4 2026).

Notes

  • Empty cells mean the mechanism is not bound to a specific clause of that profile. A cell with — means the regulation doesn't require this control — Regulus may still emit it for defence in depth.
  • Profiles bind mechanisms to specific clauses via ControlBinding. The same mechanism can be cited against different clauses in different profiles.
  • The composite profile (built from regulus.compliance.profiles: [...]) unions all bindings.
  • The composite framework (built from regulus.governance.frameworks: [...]) does the equivalent for frameworks; both join into the same audit event via regulation_clause and framework_control_id fields.