Control catalog

Every control, with the evidence behind it.

16 controls across SR 11-7, DORA, the EU AI Act, FFIEC, and ISO 42001 / SOC 2. Each entry names the clause, the objective, how the agent-session record supports it, the artifact an examiner can open, and what your organization still attests.

A reference mapping to evidence you produce — not a certification. The opinion belongs to your auditor.

Request a demo See the evidence artifacts

Catalog at a glance

16controls · 6 frameworks

Implementation evidence: 8
Monitoring signal: 3
Verifier-checkable: 4
Needs your sign-off: 1

States describe the kind of evidence that exists. None of them says a control is met.

01/SR 11-7 · Federal Reserve & OCC

Model risk management

Inventory, conceptual soundness, ongoing monitoring, and senior sign-off, applied to the agents that write your code.

SR11-7.INV-01Implementation evidence

Model Inventory & Origin Lineage

SR 11-7 Section III.A: Comprehensive model inventory and component tracking

Control objective: Maintain an accurate, updated inventory of all models, versions, and downstream autonomous agents deployed in development.
How the session record supports it: agent_activity.v1 records can include model, connector, session, and repository identifiers.
Evidence artifact: agent_activity.v1 / session_provenance
Audit chapter: Surface & Inventory
Suggested reviewer: Head of Model Risk (MRM)
Reference in code: app/lib/portal/agent-activity.ts

SR11-7.SND-02Monitoring signal

Conceptual Soundness & Intent Trace

SR 11-7 Section III.B: Conceptual soundness evaluation and design validation

Control objective: Verify that autonomous agent actions adhere strictly to stated specifications and do not execute unexpected code paths.
How the session record supports it: Intent Registry groups intent bundles by content hash. Agent Run Audit projects spec-trace evidence from scoped records.
Evidence artifact: Intent Registry / Agent Run Audit Spec Trace
Audit chapter: Spec Trace & Intent
Suggested reviewer: Head of Model Risk (MRM)
Reference in code: app/lib/portal/intent-registry.ts

SR11-7.MON-03Monitoring signal

Outcomes Analysis & Ongoing Monitoring

SR 11-7 Section III.C: Ongoing monitoring, benchmarking, and outcome analysis

Control objective: Continuously measure agent reliability, error rates, prompt churn, and cognitive debt across development teams.
How the session record supports it: Developer reports project team-scoped cognitive-debt signals and weekly trends. The portal reports signal confidence and evidence counts.
Evidence artifact: Cognitive-Debt Manager Report / agent_trust_score.v1
Audit chapter: Anomalies & Reliability
Suggested reviewer: VP of Software Engineering
Reference in code: app/lib/portal/server.ts

SR11-7.GOV-04Needs your sign-off

Model Risk Officer Threshold Sign-Off

SR 11-7 Section IV: Governance, validation policies, and senior risk sign-off

Control objective: Ensure senior model risk owners review and sign off on policy thresholds governing autonomous code generation.
How the session record supports it: Portable Model-Risk Threshold Signoff receipts (model_risk_threshold_signoff_receipt.v1) capture policy hash, signer identity, and review cadence.
Evidence artifact: model_risk_threshold_signoff_receipt.v1
Audit chapter: Governance Signoff
Suggested reviewer: Head of Model Risk (MRM)
Reference in code: packages/verifier-portal-client/src/model-risk-signoff.ts

Framework mapping

SR 11-7 model-risk governance

Documented intent, review, and exceptions, with risk monitored over time — not at one approval point.

Evidence produced

Intent Fidelity events
Requirement-to-change trace
Approvals & governance audit
PR merge evidence
Evidence package export
Cognitive-debt manager report
Agent Trust Score
Threshold sign-off receipts

Still to validate

Model-inventory linkage, independent validation, and model-risk owner signoff in your environment.

02/DORA · EU supervisory authorities

Digital operational resilience

ICT change control, third-party tool boundaries, and fail-closed behavior when review is unavailable.

DORA.ICT-01Verifier-checkable

Pre-Production Change Blast Radius

DORA Article 9: ICT Risk Management & Change Control Framework

Control objective: Identify and analyze the operational impact of AI-generated code changes prior to merging into codebase branches.
How the session record supports it: Blast Radius Snapshot computes touched routes, exported symbols, CLI commands, and caller dependencies on read from session diffs.
Evidence artifact: Blast Radius Snapshot / Provenance Query API
Audit chapter: Side Effects & Blast Radius
Suggested reviewer: Lead Application Security Architect
Reference in code: app/lib/portal/blast-radius.ts

DORA.TPRM-02Implementation evidence

Third-Party AI Tool Egress & Boundary Gate

DORA Article 28: Management of ICT Third-Party Risk & Egress Controls

Control objective: Prevent uncontrolled network egress, third-party data leakage, or unapproved API interactions by coding agents.
How the session record supports it: Bring Your Own Cloud (BYOC) readiness checks declared boundary and egress evidence. It does not deploy or validate a customer environment.
Evidence artifact: BYOC Deployment Readiness / Secret Redaction Proof
Audit chapter: Security & Boundaries
Suggested reviewer: Chief Information Security Officer (CISO)
Reference in code: packages/verifier-portal-client/src/byoc-deployment-readiness.ts

DORA.RES-03Implementation evidence

Fail-Closed Operational Resilience

DORA Article 11: Response & Recovery with Fail-Safe Controls

Control objective: Guarantee that unreachable policy servers or unverified agent payloads immediately fail closed rather than allowing uninspected execution.
How the session record supports it: The runtime approval provider fails closed on timeout or transport error. The portal records blocked actions when configured.
Evidence artifact: Runtime Approval Provider / Risk-Prevented Ledger
Audit chapter: Side Effects & Governance
Suggested reviewer: Lead Application Security Architect
Reference in code: app/lib/portal/runtime.ts

Framework mapping

DORA (EU operational resilience) ICT change & third-party risk

Controlled changes, approvals that deny when review is unavailable, and a clear view of reliance on outside agents.

Evidence produced

Governance-path docs (local, CI, worker)
Portal approval bridge
Audit records
Evidence packages
Connector proof plans & receipts
Architecture boundary statement
Cloud-boundary docs
Deployment readiness checks

Still to validate

Full deployment automation, customer-environment validation, live connector proof, and resilience runbooks.

03/EU AI Act · European AI Office

Logging, oversight & documentation

Automatic record-keeping, human oversight with rationale, technical documentation, and data-leak defenses.

EUAI.ART11-01Verifier-checkable

Technical Documentation & Architecture Specs

EU AI Act Article 11: Technical Documentation Requirements

Control objective: Maintain comprehensive, verifiable documentation detailing system architecture, capabilities, and data provenance.
How the session record supports it: Capability manifests declare permissions, event families, and trust levels. Regulated evidence bundles bind reviewed artifact digests.
Evidence artifact: Compliance Evidence Pack / Capability Manifests
Audit chapter: Surface & Capabilities
Suggested reviewer: VP of Software Engineering
Reference in code: docs/compliance-evidence-pack.md

EUAI.ART12-02Implementation evidence

Automated Tamper-Evident Event Logging

EU AI Act Article 12: Automatic Record-Keeping of Operations

Control objective: Automatically capture complete, immutable logs of all agent tool calls, model inferences, inputs, and outcomes throughout operations.
How the session record supports it: The event ledger stores signed agent_activity.v1 records. Evidence export and provenance rows use append-only hash chains.
Evidence artifact: agent_activity.v1 Ledger / Append-Only Export Chain
Audit chapter: Audit Ledger & Chains
Suggested reviewer: Chief Information Security Officer (CISO)
Reference in code: app/api/verification-portal/ingest/route.ts

EUAI.ART14-03Implementation evidence

Human-in-the-Loop Oversight & Rationales

EU AI Act Article 14: Human Oversight & Interventions

Control objective: Enable effective human intervention to gate, inspect, approve, or halt autonomous actions with documented reviewer rationale.
How the session record supports it: Runtime sensitivity can hold specified tool calls. The approval ledger records reviewer decisions and available rationale.
Evidence artifact: Approval Decision Records / Deviations Review Queue
Audit chapter: Human Decisions & Rationale
Suggested reviewer: Head of Model Risk (MRM)
Reference in code: app/verification-portal/approvals/page.tsx

EUAI.ART15-04Implementation evidence

Cybersecurity & Data Redaction Defense

EU AI Act Article 15: Accuracy, Robustness, and Cybersecurity

Control objective: Protect against sensitive data leakage, credential poisoning, and unredacted customer data passing into LLM telemetry.
How the session record supports it: Deterministic regex catalog (AWS, GitHub, JWT, Stripe keys) + Shannon entropy fallback scrubs credentials before payload emission.
Evidence artifact: Redact-Before-Emit Summary / Manifest Redaction Policy
Audit chapter: Redaction & Security
Suggested reviewer: Data Protection Officer (DPO)
Reference in code: packages/verifier-portal-client/src/secret-redactor.ts

Framework mapping

EU AI Act governance concepts

Demonstrable human oversight: approvals on record, tamper-evident logs, documented risk decisions.

Evidence produced

Human approval bridge
Append-only audit log
Intent Fidelity notes
Agent Run Audit
Policy decision reasons
Evidence export

Still to validate

Legal classification, role-specific human-oversight procedures, and retention-policy alignment.

04/FFIEC · Federal Financial Institutions Examination Council

Bank IT examination

Dual control, separation of duties, and attributed SDLC code provenance.

FFIEC.SOD-01Implementation evidence

Dual-Control & Separation of Duties

FFIEC Information Security Handbook: Separation of Duties & Access Governance

Control objective: Prevent single-person execution of critical system modifications or autonomous actions by requiring independent review.
How the session record supports it: Role checks separate development and approval permissions. The approval contract can require an independent second review.
Evidence artifact: Portal Access Control / Approval Decision Ledger
Audit chapter: Human Decisions & Rationale
Suggested reviewer: Chief Information Security Officer (CISO)
Reference in code: app/lib/portal/access.ts

FFIEC.SDLC-02Verifier-checkable

Attributed SDLC Code Provenance

FFIEC Development & Acquisition Booklet: Software Development Lifecycle Integrity

Control objective: Attribute all code modifications to verified developer identities, git commits, and agent sessions with complete provenance.
How the session record supports it: Pull request (PR) evidence can bind recorded sessions, commit hashes, policy decisions, and Check Run results into one digest.
Evidence artifact: PR Merge Evidence Summary / GitHub Check Run
Audit chapter: Spec Trace & Merge Gates
Suggested reviewer: Lead Application Security Architect
Reference in code: packages/verifier-portal-client/src/pr-evidence.ts

Framework mapping

FFIEC bank technology risk

Instrumented changes trace to their agent; each gated action to an actor, a policy, and an outcome.

Evidence produced

Normalized event ledger
Policy decisions
Approval bridge records
Audit explorer
PR merge evidence & Check Runs
Merge-gate & branch protection
Connector provenance
Pre-execution policy blocks

Still to validate

IAM integration, live identity mapping, and a validated deployment inside your own cloud.

05/ISO 42001 / SOC 2 · ISO/IEC & AICPA

AI management & trust services

Continuous AI risk assessment, signed access, and verifiable change-management evidence.

ISO42001.RSK-01Monitoring signal

Continuous Agent Trust & Risk Scoring

ISO/IEC 42001 Clause 6.1: Actions to Address Risks and Opportunities in AI

Control objective: Maintain ongoing systematic assessment of AI risks across all integrated agent tools, harnesses, and repositories.
How the session record supports it: Team and agent views project agent_trust_score.v1 from scoped activity and approval records. It is not a compliance score.
Evidence artifact: agent_trust_score.v1 Chapter / Deviations Queue
Audit chapter: Anomalies & Reliability
Suggested reviewer: Head of Model Risk (MRM)
Reference in code: app/lib/portal/server.ts

SOC2.CC6-02Implementation evidence

Role-Based Authentication & Signed Ingest

SOC 2 Common Criteria 6.1: Logical Access Controls & Identity Verification

Control objective: Restrict access to telemetry and governance controls to authenticated, authorized users and secure ingest endpoints with cryptography.
How the session record supports it: Hash-based Message Authentication Code (HMAC) protects ingest and runtime approval requests. Portal queries use authenticated team scopes.
Evidence artifact: Runtime Request Signing / Portal Access Guards
Audit chapter: Security & Boundaries
Suggested reviewer: Chief Information Security Officer (CISO)
Reference in code: app/lib/portal/runtime-auth.ts

SOC2.CC8-03Verifier-checkable

Cryptographic Evidence Package Verification

SOC 2 Common Criteria 8.1: Change Management & System Integrity

Control objective: Provide verifiable, tamper-evident proof that software changes underwent authorized verification and policy checks prior to deployment.
How the session record supports it: Evidence exports use manifests and SHA-256 digests. Persisted exports record append-only chain entries.
Evidence artifact: Evidence Package Manifest / Local Verifier Transcript
Audit chapter: Audit Ledger & Chains
Suggested reviewer: VP of Software Engineering
Reference in code: packages/verifier-portal-client/src/evidence-package.ts

Framework mapping

ISO 42001 AI management system

An AI management system needs defined roles, risk controls, monitoring, and traceable records.

Evidence produced

Scope & boundary documentation
Private registry provenance
Policy packs
Cognitive-debt report
Agent Trust Score
Audit & evidence exports
Control-ownership matrix template

Still to validate

Formal AI management-system procedures, completed control-owner matrices, and an approved trust-score use policy.

Framework mapping

SOC 2 / ISO 27001 change management

The same questions as human code: authorized by whom, tested how, traceable to what record.

Evidence produced

PR merge evidence
Approvals & audit records
Agent Run Audit
Conformance tests
Check Run & merge gate
Branch-protection verifier
BYOC deployment-readiness report

Still to validate

CI build and test results, ticketing integration, and a validated deployment inside your own cloud.

AutoDevOps produces the evidence; the opinion belongs to your auditor. Customer-cloud validation of AutoDevOps itself is still open in every mapping above.

Evidence, not attestations

Eight artifacts. One system of record.

The core artifacts behind the catalog, each one something your auditor can open and verify. Controls that rely on something more specific — a sign-off receipt, a readiness report, a redaction summary — name it in their own entry above.

Normalized agent events

Each tool and model call a connector captures, in one schema across agent vendors.

SR11-7.INV-01 EUAI.ART12-02

Policy decisions

Allow, confirm, or block, recorded with the rule that fired.

DORA.RES-03 EUAI.ART14-03

Human approvals

Who approved what, and when. Portal reviews also require a rationale.

EUAI.ART14-03 FFIEC.SOD-01

Append-only audit log

Hash-chained records, so an edit is detectable when checked against a trusted hash.

EUAI.ART12-02 SOC2.CC8-03

Intent Fidelity evidence

What was asked, bound to what the agent actually built.

SR11-7.SND-02

Provenance queries

Audited answers to “which sessions touched this change?”

DORA.ICT-01

PR merge evidence

The decision record inside the PR reviewers already read.

FFIEC.SDLC-02

Evidence packages

Tamper-evident exports your auditor can verify offline.

SOC2.CC8-03

See one control’s evidence, end to end.

Pick a control from the catalog. We walk the validated AWS path from agent session to policy decision, approval, and the evidence package that answers it.

Request a demo GitHub

Building a connector or wiring approvals? Read the docs.