review
Periodic review of an AI worker's behaviour and evidence.
KYE™ Assurance Card Profile™ · v1.0
From system cards to runtime assurance.
From system card to runtime assurance. Signed evidence. Scoped to your tenant. Replayable in 60 seconds.
Assurance Card Profile: side by side
Demo dataYou learn: How the options differ, criterion by criterion. You can: Pick a column to light it.
| Models | id, version, supplier, licence, checksum | self_attested · third_party_attestation · signed_evidence · registry_check |
|---|---|---|
| Datasets | id, version, licence, checksum, data classes | same |
| Tools | id, version, supplier | same |
| Suppliers | id, verification flag, method | same |
| Licences | SPDX identifier or named licence | registry_check |
| Hardware | id, vendor, model, attestation ref | signed_evidence |
How this widget is built
- Demonstrates
- Comparison
- Components
- The options this page compares
- The criteria it compares them on
- Flow
- Each criterion across the options, as the page sets it out. Architecture diagram
- Used on
- /assurance-card/: On this page, after its opening.
KYE™ Assurance Card Profile™ turns KYE™ runtime evidence into a living assurance record for AI agents, models, tools, workflows, connectors and delegated systems. KYE™ makes assurance executable.
Plain take
Plain take
From card to proof. Signed. Scoped. Replayable.
- One card per system. One pack per call.
- Auditors get OSCAL. Same data. Half the time.
- Regulators get scoped evidence. No raw data.
- You keep your keys. Always.
Why a new layer
Model cards are not enough.
Defence-grade AI assurance work (Alan Turing Institute / Accenture / UK MoD) makes the case directly: model cards are insufficient where models are integrated with platforms, where assumptions change and where multiple models interact. Agentic systems need more than model cards. They need authority cards, runtime decisions and evidence trails.
KYE™ Assurance Card Profile™ sits above the runtime KYE™ stack. It turns the operating-model journey into a living lifecycle record, automates review-cycle triggering against the authority graph, and enforces the human-involvement plan as a runtime policy gate — not a documentation requirement.
1 · 8-stage lifecycle
Design → pilot → deploy → monitor → review → decommission.
Each stage emits a signed event onto KYE™ Signal Bus™ and is recoverable from the audit ledger.
01
DesignKYEAssuranceCard drafted; intended use, prohibited uses, owner, accountable entity declared.
02
PilotCard moves to pilot. Operating-model adoption pack bound; runtime evidence begins flowing.
03
DeployStatus moves to controlled_production / production; cascade-revocation hooks armed.
04
MonitorKYEAssuranceReviewCycle watches authority-graph + provenance + risk-state for triggers.
05
Incident reviewTrigger incident instantiates a review without operator intervention.
06
Scope-change reviewTrigger scope_change on capability / gate / commit-boundary updates.
07
Retention reviewTrigger retention on data / evidence retention windows.
08
DecommissionKYEDecommissioningPlan orchestrates revoke · quarantine · rotate · archive · notify · remove.
2 · Live assurance-card preview
Owner. Authority. Human involvement. Provenance. Evidence.
Every claim on a KYE™ Assurance Card™ resolves to a runtime artefact a third party can replay offline against a published JWKS.
KYE™ Assurance Card™ · preview
Finance Payment Preparation Agent
pilot
3 · KYE™ Human Involvement Plan™
Where authorised humans must review, approve, limit or override.
Six involvement types · nine lifecycle stages. Bypass attempts are denied at runtime and emit a signed kye.human_involvement.bypass_denied event.
approve
Required approval before commit-level action.
limit
Constrain scope or output before execution.
direct
Provide direction or guardrails the AI worker must follow.
influence
Set policy or framing that shapes interpretation.
override
Reverse or change a runtime decision after the fact.
4 · Provenance & supply chain
Models. Datasets. Tools. Suppliers. Licences. Verified.
Every dataset, model, tool, supplier, licence and hardware lineage referenced by an assurance card is bound as KYEProvenanceEvidence. Verification status: verified / partially_verified / unverified / failed. Verification failure emits a signed kye.provenance.verification_failed event onto Signal Bus™.
| Element | What is recorded | Verification methods |
|---|---|---|
| Models | id, version, supplier, licence, checksum | self_attested · third_party_attestation · signed_evidence · registry_check |
| Datasets | id, version, licence, checksum, data classes | same |
| Tools | id, version, supplier | same |
| Suppliers | id, verification flag, method | same |
| Licences | SPDX identifier or named licence | registry_check |
| Hardware | id, vendor, model, attestation ref | signed_evidence |
5 · Review cycle (interactive)
Pick a trigger — we'll show what it produces.
Front-end illustration only — no data leaves the page. Real implementations call POST /v1/assurance-review-cycles/{cycle_id}:trigger.
When does it fire?
On the cadence declared by scheduled_frequency (weekly / monthly / quarterly / annual).
Reviewer roles: ai_governance_lead. Evidence captured: reviewer_identity, decision, reason_code, timestamp, state_snapshot, decision_map.
When does it fire?
When the bound KYEEntityAuthorityRecord adds or removes a capability, or when the KYEAuthorityGate set changes membership.
Reviewer roles: ai_governance_lead + security_admin. Evidence captured: scope_diff, decision_map, state_snapshot.
When does it fire?
When an incident signal is recorded (kye.compromise.reported, kye.agency_drift.detected, etc.) for a descendant of the assurance-card subject.
Reviewer roles: security_admin + ai_governance_lead. Outcome options: approve / deny / quarantine / escalate. SLA: 240 minutes.
When does it fire?
When KYEProvenanceEvidence.models[*].version changes.
Reviewer roles: ai_governance_lead + data_steward. Evidence captured: model_diff, evaluation_refs, performance_metrics.
When does it fire?
When KYEProvenanceEvidence.suppliers[*].verified changes from true to false, or a new supplier is added.
Reviewer roles: procurement + security_admin. Evidence captured: supplier_diff, verification_method, signed_evidence.
When does it fire?
When a model / dataset / tool licence changes.
Reviewer roles: legal. Evidence captured: licence_diff, registry_check.
When does it fire?
When the data / evidence retention window approaches expiry.
Reviewer roles: data_steward + legal. Outcome options: archive / extend / delete / decommission.
When does it fire?
When the assurance card transitions to deprecated or revoked, or the bound KYEDecommissioningPlan is executed.
Off-boarding: revoke_authority → quarantine_credentials → rotate_keys → archive_evidence → notify_owner → notify_supplier → update_catalog → remove_from_runtime. Cascade scope: per cascade_revocation.scope.
6 · Schemas (Apache 2.0, public mirror)
Six normative objects. Validated in CI.
What it is. Why it matters. What to do next.
assurance-card-profile.json— profile manifestassurance-card.json— living lifecycle assurance recordhuman-involvement-plan.json— KYE™ Human Involvement Plan™provenance-evidence.json— supply-chain + model / dataset / tool lineageassurance-review-cycle.json— scheduled + event-triggered reviewsdecommissioning-plan.json— retention + off-boarding + cascade revocation
7 · Open / paid boundary
The contracts are open. The runtime assurance engine is paid.
One sentence: a signed, replayable proof of decision.
Open source
Open
- Assurance Card Profile™ schema
- KYE™ Assurance Card™ schema
- KYE™ Human Involvement Plan™ schema
- Provenance Evidence schema
- Assurance Review Cycle schema
- Decommissioning Plan schema
- Reason-code dictionary · Signal Bus™ event names
- Sample assurance cards · basic conformance fixtures
Commercial track
Paid
- KYE™ Assurance Card Builder™
- KYE™ Assurance Card Library™
- KYE™ Human Involvement Planner™
- KYE™ Provenance & Supply Chain Evidence App™
- KYE™ Assurance Review Scheduler™
- KYE™ Governed Use Case Library™
- Sector packs (defence, public-sector, payments, health)
- Runtime-evidence binding + review-automation engines
- BYOC / on-prem assurance deployment
Where to go next
Adjacent reading.
What it is. Why it matters. What to do next.
Ready to see your AI agents flagged?
Start in shadow mode. We’ll deliver your first Evidence Pack™ in 4–8 weeks.
Canonical KYE™ surfaces referenced on this page: Authority Graph™ · Evidence Pack™ · KYE Protocol™.