How we run our own management system, with agents
Most ISO 42001 pages explain the standard. This one shows a management system running: Sentry AI's own, the real counts from our register as at 7 October 2026, what agents do, what people still decide, and the one gate we have not passed.
The short answer
Sentry AI runs its information security management system with agents doing the clerical work and people making every decision. As at 7 October 2026 the register holds 27 assessed risks, 86 adopted controls, a decision on all 93 Annex A items (85 addressed, 8 excluded), 51 evidence entries and 30 recorded decisions, and six of seven readiness gates are passed. ISO/IEC 42001 runs on the same control set. Neither programme is certified yet: the last gate needs every finding closed and the loop shown to run.
The register, as it stands
Counts read from our own register on 7 October 2026. These are the ISO/IEC 27001 information security management system, which is the foundation our ISO 42001 work is built on: the two standards share one structure, so the controls, document control, internal audit and management review carry straight across.
Annex A applicability decisions made
85 controls addressed, 8 excluded with a written justification, none left undecided.
Controls adopted
15 implementations already operating on real systems, of 19 recorded.
Risks assessed
Every one with a residual assessment. No high risk left untreated or unaccepted.
Decisions recorded
Each one dated and attributed to a named person, at the time it was made.
Evidence entries
Collected through the connectors we already use, with five awaiting a human check.
Test definitions
Repeatable checks of whether a control works, with 9 assessments run so far.
Six of seven gates passed, and the honest one
Our engine checks the management system against seven readiness gates in the order an auditor would. The first six test whether the system exists. The seventh tests whether it runs: every finding and every exception closed, with the loop shown to work. That gate is not passed, and it is the right one to be stuck on. It takes operating time, not more documents, which is why most programmes that stall, stall here.
| Gate | Status | What it checks |
|---|---|---|
| Context and scope | Passed | Organisation profile, scope and risk methodology written down. |
| Leadership | Passed | 4 security objectives and 4 approved governed documents. |
| Risk and Statement of Applicability | Passed | 27 risks assessed; all 93 Annex A items decided. |
| Support | Passed | The workforce recorded and evidence collection running. |
| Implementation | Passed | Adopted controls operating on real systems. |
| Evaluation | Passed | Tests defined, internal audit and management review scheduled. |
| Improvement and readiness | Not yet | Every finding and exception closed, with the loop demonstrably run. |
Who does what: engine, agents, people
The question every small company asks about ISO is where the hours come from. Our answer is that agents take the clerical hours and people keep the decisions.
What the engine owns
The state. Every register (risks, controls, assets, suppliers, decisions, evidence, tests, findings) is a file in a version-controlled repository, and every change is written to an append-only, hash-chained ledger. Validation lives here too: the engine refuses an action the methodology does not allow, such as closing a finding that has no action against it.
What agents do
The clerical work, which is most of it. Claude drafts and revises the governed documents, collects evidence through the same connectors we use to run the business, runs the test definitions, raises findings when practice and paper part ways, and keeps the registers current as systems change.
What people decide
Everything an auditor will ask a person about. Approving a document, accepting a risk, excluding an Annex A control, closing a finding: each is a decision recorded against a named human, at the time. Agents propose; people decide. The standard requires accountable people, and an auditor interviews them.
Where the agents themselves are governed
Our own platform is in scope. Every action an agent can take in the Sentry AIOS hub is classed AUTO, APPROVE or BLOCK by one written policy, enforced in one place, so a new tool cannot arrive ungoverned. The management system found the need for that itself, as a finding against our own approval gate.
What it taught us about your programme
Three things we would tell anyone starting ISO 42001. Start with the inventory: you cannot scope or risk assess what nobody has listed. Record decisions at the time: an approval written up afterwards is the first thing an auditor doubts. And let the system find faults in itself: our most useful findings were raised against our own tooling, not by a consultant reading a policy. The free readiness check asks the twenty questions we had to answer.
Part one of a series
We will publish the register again as the programme moves: the seventh gate, the internal audit, the management review and the certification audits, with the counts as they stand each time.
Frequently asked questions
Can a small company run an ISO management system?
Yes. The standards scale to the organisation: scope, risks and controls are sized to what you actually do. We are a small team and run a full information security management system, with every Annex A applicability decision made, by keeping the registers in files and letting agents do the clerical work while people make the decisions.
Can AI agents run an ISO 27001 or ISO 42001 management system?
Agents can do most of the work and none of the accountability. In ours, agents draft documents, collect evidence through the connectors we already use, run the tests and raise findings. Every decision, approval and acceptance of risk is recorded against a named person, because the standards require accountable people and an auditor will interview them.
Is Sentry AI ISO 42001 or ISO 27001 certified?
Not yet. Both programmes are in progress on one control set, and we will say so on our security page the day an accredited body certifies either, and not before. This page shows the management system as it actually stands, including the gate we have not passed.
What is the hardest part of running the management system?
Closing the loop. Writing policies and adopting controls is the easy half. The last readiness gate needs every finding and every exception closed with the loop demonstrably run, and that takes operating time, not documents.
How does ISO 27001 help with ISO 42001?
Both use the same harmonised structure: context, leadership, planning, support, operation, evaluation and improvement. Document control, internal audit, management review and the evidence habit carry straight across. ISO 42001 adds the AI-specific work on top: the AI inventory, AI system impact assessments and the Annex A controls for the AI lifecycle.
Want to know how far you are from Stage 2?
Take the free readiness check first. If the gaps are real, a short call will show you what building the management system involves for a business your size.
Ask an AI about this page
Opens the assistant with this page loaded: read it, summarise it, cite it.