Skip to main content

ISO/IEC 42005:2025 evidence pack

ISO/IEC 42005:2025 is the AI Impact Assessment guidance that maps to ISO 42001 controls. Ujex emits a structured evidence pack covering the §7.2 through §7.8 requirements.

Calling the export

const pack = await fns.httpsCallable('auditExportISO42005')({
from: '2026-04-01T00:00:00Z',
to: '2026-07-01T00:00:00Z',
});

Shape

{
"pack_kind": "iso-iec-42005:2025",
"range": {"from": "...", "to": "..."},
"generated_at": "...",
"agents": [{"agent_id", "display_name", "created_at", "disabled"}],
"counts": {
"tool_call": 12000,
"approval": 134,
"memory_write": 4500,
"outbound_email": 2300,
"inbound_email": 9800,
"policy_decision": 23,
"human_override": 12,
"governor_block": 9,
"other": 88,
"denied": 47,
"error": 18
},
"samples_by_action_type": { "tool_call": [{seq, hash, action, ts}, ...] },
"evidence_map": {
"iso-42005-§7.2 Agent inventory": {...},
"iso-42005-§7.4 Human oversight": {...},
"iso-42005-§7.5 Risk controls": {...},
"iso-42005-§7.6 Data handling": {...},
"iso-42005-§7.8 Continuous monitoring":{...}
}
}

The evidence_map is the auditor-facing artefact. Each entry names the ISO 42005 requirement, the Ujex query/source that satisfies it, and the count of evidence rows in the window.

What the auditor sees vs Ujex requires

Requirement (§7.x)Audit queryWhere it's emitted
§7.2 Agent inventoryagents collectionAgent registry
§7.4 Human oversightaction_type IN (approval, human_override)Approvals + Admin
§7.5 Risk controlsaction_type IN (policy_decision, governor_block) + outcome=deniedBudgets + Approvals policy
§7.6 Data handlingaction_type IN (memory_write, inbound_email, outbound_email)Memory + Inbox
§7.8 Continuous monitoringHash-chain + verifyChain scheduleAudit subsystem

Pairing with §7.7 (model performance)

ISO 42005 §7.7 wants ongoing model performance evidence. Ujex doesn't collect model accuracy metrics directly — pair with Langfuse / Phoenix / Arize and reference their evidence pack alongside Ujex's.