// compliance

MoorAI & ISO/IEC 42001: content-free evidence for an AI Management System

ISO/IEC 42001 asks two things of you: manage the risk of the AI you run, and be able to prove you did. Most tooling proves the second by inspecting, and storing, the prompts and payloads you were trying to govern. MoorAI produces that proof a different way: content-free, on the device. Below, MoorAI's shipped controls are mapped to the parts of ISO 42001 they speak to.

Where MoorAI fits: ISO 42001 is about a management system for AI: a lifecycle you operate and a set of risks you treat and evidence. MoorAI generates that evidence where the AI runs, on the developer's endpoint, and by default without a prompt leaving the machine.

5.4AI system lifecycle processes — a system of record at the endpoint

AI Agent Gateway · AIBOM · shadow-AI discovery

Clause 5.4 expects defined, operated processes across the AI system lifecycle. On a developer's machine, the AI system is the coding agent and everything it is wired to. MoorAI gives that lifecycle a system of record, on the device:

The on-device AI Agent Gateway is one chokepoint per MCP tool-call — every invocation an agent makes passes through a single, observable point rather than fanning out unwatched.
The AI Bill of Materials (AIBOM) is a live inventory of the agents, models, and MCP servers actually present — the running composition of the AI system, not a stale diagram.
Shadow-AI app and browser discovery enumerates the agents and AI surfaces on each enrolled machine, so the lifecycle covers what is really running, including the unmanaged installs.

Together these are a system of record across the agent lifecycle at the endpoint (discovery, inventory, and a single tool-call chokepoint), all built from configuration metadata on the device.

8Operation — AI risk assessment & treatment at the moment of action

coach / alert / block · JIT elevation · assurance score · model allow-list

Clause 8 is where the management system meets operation: you assess AI risk and you treat it. MoorAI treats risk at the one place it is unambiguous — the moment an action is about to happen:

Per-action enforcement (coach, alert, or block) before a prompt reaches the agent or an MCP tool-call runs. Risk treatment applied at the action, not reconstructed after the fact.
Just-in-time elevation for privileged steps: an agent that needs to do something sensitive gets a scoped, time-boxed grant instead of standing permission.
A per-agent assurance score that expresses how much an agent has earned, so treatment can be proportionate to assessed risk.
A model-endpoint allow-list that bounds which models an agent may reach — a treatment control on where inference is allowed to go.

Each of these is a risk-treatment decision taken deterministically, at the moment of action, on the device.

Annex AReference controls — evidence you can hand an auditor without handing over the content

Event Flow / data-lineage · signed decisions · compliance packs · A.2.5 system security & data governance

Annex A is ISO 42001's catalogue of reference controls — things like system security and data governance (e.g. A.2.5). MoorAI's answer to these is deliberately content-free, so the evidence is safe to disclose as-is:

A content-free Event Flow / data-lineage view: which class of data moved through which agent, by category and keyed one-way hash, against the policy that allowed or blocked it — never the data itself.
Cryptographically signed agency decisions: each enforcement decision recorded as a tamper-evident record of what was allowed, under which policy, on which device.
Compliance packs that map live, per-tenant evidence — MoorAI maps its controls to ISO 27001/42001 (alongside OWASP LLM Top 10, NIST CSF, NIST AI RMF, SOC 2, and the EU AI Act).

Because every artifact is drawn from redacted, content-free signals by default, you can hand an auditor the evidence of how you govern AI — without also handing over a copy of what anyone actually wrote. One exception: an administrator can turn on a capture tier, which does store the matched text it was enabled to collect.

Evidence without egress. The through-line across Clause 5.4, Clause 8, and Annex A is the same: the control runs where the prompt is typed, so the proof exists as signed, content-free records. You demonstrate a managed, evidenced AI system — without routing prompts or payloads through a vendor's cloud to do it.

Informational, not a certification. This page maps MoorAI's shipped controls to ISO/IEC 42001 clauses and Annex A reference controls to help teams assemble evidence; it does not constitute certification, and it is not a legal or conformity assessment. ISO/IEC 42001 certification is issued by accredited bodies against a full audit of your AI Management System. Verify current clause references against the standard itself.

See how MoorAI produces this evidence →

MoorAI — content-free runtime guardrails for AI agents, apps and APIs. Governance without surveillance. See also MoorAI vs Salt Security and the five questions every AI-governance buyer is being asked.