This is the register I would publish and hold teams to. The test for every entry is the same: name the mechanism that enforces it and the evidence it produces. A standard that fails that test is an aspiration, and aspirations do not survive an audit.
| ID | Standard | Enforced by | Evidence produced |
|---|---|---|---|
| AS-01 | Every agent is registered. Identity, accountable owner, purpose, tier, data classes, tools, expiry. | Gateway rejects unregistered agent identities | Registry record, ownership attestation |
| AS-02 | Autonomy is tiered. R0 to R4 assigned by reversibility, blast radius, data sensitivity and regulatory scope. | Tier is an attribute of the registry entry; controls attach to the tier | Risk assessment, tier decision record |
| AS-03 | No standing credentials. Agents receive short-lived, narrowly scoped tokens at the point of use. | Tool broker performs the exchange; secret scanning in CI | Token issuance log with scope and TTL |
| AS-04 | Delegated authority never exceeds the principal. An agent acting for a person inherits that person's entitlements, no more. | Policy decision point evaluates the propagated principal scope per call | Authorization decision log |
| AS-05 | One authoritative policy system, no ungoverned paths. Policy is authored and versioned once, then enforced at the gateway, tool broker, memory service, retrieval layer, workflow engine, model gateway and egress proxy. | Deny-by-default egress plus conformance tests proving every enforcement point evaluates the same bundle | Policy version history, enforcement-point coverage report, exception register |
| AS-06 | Writes are idempotent and reversible. Every mutating tool declares an idempotency key and a compensating action. | Tool contract test in CI; broker rejects non-conforming tools | Contract test results, rollback rehearsal record |
| AS-07 | Budgets are bound at start. Steps, tokens, spend and wall clock declared per task and enforced at runtime. | Gateway counters; breach halts the run and notifies the owner | Per-task cost and step telemetry |
| AS-08 | Untrusted content is screened both ways. Retrieved documents, tool responses and peer messages are treated as hostile input. | Input and output guardrail services on the gateway path | Screening verdicts, blocked-event log |
| AS-09 | Memory carries provenance. Source, writing task, owning scope and expiry on every durable entry. | Memory service schema; writes without provenance are rejected | Memory audit trail, deletion receipts |
| AS-10 | Tools and skills are signed and pinned. No unpinned remote definitions, no unreviewed public MCP servers. | Private registry with signature verification at load time | Software bill of materials, scan results |
| AS-11 | Evidence is written before the action. Intent, decision and inputs are logged first; failed logging fails the action. | Gateway write-ahead to the append-only decision log | Tamper-evident decision log |
| AS-12 | Releases pass published thresholds. Eval pass rate, injection resistance, cost per task and latency, agreed with the business. | CI gate blocks promotion below threshold | Eval report attached to the release |
| AS-13 | Production is continuously evaluated. Sampled runs scored, drift tracked, judge models re-calibrated against humans. | Scheduled sampling jobs with alerting on trend breaks | Quality trend, judge agreement rate |
| AS-14 | Stop works. Per-agent, per-tool and global kill switches, tested on a schedule. | Control plane feature flags with independent operator access | Kill-switch test log, incident drill report |
| AS-15 | Hard limits are deterministic. Tool and parameter allow-lists, value and rate caps, tenant and data boundaries, egress control and sandboxing are code, not model behaviour. Screening supplements them and never replaces them. | Enforcement points reject out-of-envelope calls before execution | Envelope definition, rejected-call log, egress policy |
Maintain one control set and map it outward, so an auditor's question is answered from the same pipeline that runs the platform. Two cautions before you use this. ISO/IEC 42001 is an AI management system standard describing organizational capability, not a technical agent architecture, so these rows connect a technical control to the management-system control it produces evidence for. And EU AI Act duties differ by role, so the provider or deployer column matters as much as the article number.
| ID | Control | ISO/IEC 42001 | NIST AI RMF | AI 600-1 area | EU AI Act | Role | IMDA | OWASP |
|---|---|---|---|---|---|---|---|---|
| AS-01 | Agent registration | A.3.2 · A.4.2 · A.6.2.2 | GV 1.3 · GV 2.1 · MP 1.1 | Accountability structures | Art. 11 · 16 · 26(1) | P + D | D1 · D2 | ASI03 · ASI10 |
| AS-02 | Tiered autonomy by use case and action | A.5.2 · A.5.4 · A.6.2.2 | MP 1.1 · MP 5.1 · MS 1.1 | Context and risk framing | Art. 9 | P | D1 | ASI03 · ASI08 |
| AS-03 | No standing credentials | A.6.2.6 + 27001 A.5.15–18 | GV 6.1 · MG 2.2 | Access and credential controls | Art. 15 | P | D3 | ASI03 · ASI05 |
| AS-04 | Delegated authority never exceeds the principal | A.3.2 · A.6.2.6 | GV 1.3 · MG 2.2 | Roles and delegation | Art. 14 · 26 | D | D2 · D3 | ASI03 · ASI09 |
| AS-05 | One authoritative policy system, no ungoverned paths | A.2.2 · A.6.2.6 · A.9.2 | GV 1.2 · MG 1.3 | Policy and process | Art. 15 | P | D3 | ASI02 · ASI10 |
| AS-06 | Idempotent and reversible writes | A.6.2.4 · A.6.2.6 | MG 2.3 · MG 4.1 | Incident response and recovery | Art. 14(4) · 15 | P | D3 | ASI02 · ASI08 |
| AS-07 | Budgets bound at task start | A.4.3 · A.6.2.6 | MG 2.2 · MS 2.6 | Resource and reliability | Art. 15 | P | D3 | ASI08 |
| AS-08 | Untrusted content screened both ways | A.6.2.4 · A.7.4 | MS 2.7 · MG 2.2 | Security and resilience testing | Art. 15 | P | D3 | ASI01 · ASI06 · ASI07 |
| AS-09 | Memory carries provenance | A.6.2.6 · A.7.2 · A.7.5 | MP 4.1 · MS 2.8 · MG 4.1 | Provenance and data lineage | Art. 10 · 12 | P | D3 | ASI06 |
| AS-10 | Tools and skills signed and pinned | A.9.2 · A.9.3 | GV 6.1 · MP 4.1 | Third party and supply chain | Art. 15 · 25 | P | D3 | ASI04 |
| AS-11 | Evidence written before the action | A.5.3 · A.6.2.6 | MS 1.1 · MG 4.1 | Monitoring and record-keeping | Art. 12 · 19 · 26(6) | P + D | D2 · D3 | ASI09 · ASI10 |
| AS-12 | Releases pass published thresholds | A.6.2.4 · A.6.2.5 | MS 2.3 · MS 4.2 · MG 1.3 | Pre-deployment testing | Art. 9 · 15 · 17 | P | D1 · D3 | ASI01 · ASI08 |
| AS-13 | Production continuously evaluated | Cl. 9.1 · A.6.2.6 | MS 2.4 · MG 4.1 | Post-deployment monitoring | Art. 72 · 26(5) | P + D | D3 | ASI06 · ASI08 |
| AS-14 | Stop works | Cl. 10.2 · A.6.2.6 | MG 2.3 · MG 4.1 | Incident containment | Art. 14(4)(e) · 26 | P + D | D2 · D3 | ASI08 · ASI10 |
| AS-15 | Hard limits are deterministic | A.6.2.4 · A.6.2.6 | MG 2.2 · MS 2.7 | Technical safeguards | Art. 15 | P | D3 | ASI02 · ASI03 · ASI05 |
If you are standing up an agent platform, tightening the controls on one you already have, or preparing for an audit that now includes agents, I am happy to look at it with you.
Tell me where you are with agents and what you are trying to make safe. I reply to every enquiry within two business days.