Skip to main content

AI Product (AIPS) Model Reference

This document describes the canonical AIPS 0.1 model.

AIPS defines an AI Product as an immutable, governed, semantic artifact that is:

  • Behaviorally classified
  • Risk-evaluated
  • Deployment-bound
  • Marketplace-addressable
  • Trust-wrapped (via DPP)

AIPS separates:

  • Core Regulatory Kernel
  • Deployment Specification (AIPDS)
  • Semantic Descriptor (AIPROD)
  • Trust Envelope (DPP)

This reference focuses on the Core AIProduct model.


1. Conceptual Overview


2. AIProduct (Core Class)

Canonical class representing a governed AI Product.

An AI Product must be:

  • Deployable
  • Self-describing
  • Behaviorally classified
  • Risk evaluated
  • Version immutable
  • Provenance-aware

If these are missing, the artifact is likely only an AI Asset, not a full AI Product.


aips:label

FieldValue
Identifieraips:label
Domainaips:AIProduct
Rangexsd:string

Human-readable name.


aips:description

FieldValue
Identifieraips:description
Domainaips:AIProduct
Rangexsd:string

Consumer-facing explanation.


aips:purpose

Captures business intent and regulatory alignment context.


aips:aiProductOwner

| Range | prov:Agent or IRI |

Accountable authority.

Critical for governance and trust.


3. Capability Modeling

aips:hasCapabilityType

Range: aips:CapabilityType

Enumeration includes:

  • Predictive
  • Generative
  • Agentic
  • Symbolic
  • Hybrid

At least one required.


4. Lifecycle Modeling

aips:hasLifecycleStage

Enumeration:

  • Experimental
  • Beta
  • Production
  • Deprecated
  • Retired

Exactly one required.


5. Behavioral Summary (Mandatory)

Behavioral modeling is ontology-aligned.

All behavioral properties are IRIs referencing:

https://kivanura.org/spec/aips/agent#

BehavioralSummary

PropertyRange
topologyScopeagent:Agent (IRI)
authorityScopeagent:Agent (IRI)
autonomyLevelagent:AutonomyLevel (IRI)
controlModeagent:ControlMode (IRI)

Behavior is not free text.

It is computable classification.


6. EffectiveRiskTier (Mandatory)

EffectiveRiskTier is:

  • Immutable
  • Provenance-aware
  • SHACL-validated
  • DPP-required

Structure

PropertyRange
tierR0–R4
computationTimestampxsd:dateTime
computationMethodxsd:string
escalationRequiredxsd:boolean

Rules:

  • If tier = R4 → escalationRequired MUST be true.
  • If tier ≠ R4 → escalationRequired SHOULD be false.

Risk tier is a computed artifact, not manual labeling.


7. Separation of Concerns

AIProduct itself does NOT contain:

  • Deployment infrastructure details
  • Marketplace publish projection
  • Cryptographic signature
  • Revocation chain

Those belong to:

  • AIPDS
  • AIPROD
  • DPP

8. AIPROD (Semantic Descriptor)

Defines:

  • Immutable semantic contract
  • Inputs
  • Outputs
  • Behavioral declarations
  • Risk exposure linkage

AIPROD is:

  • Marketplace-facing
  • Version-immutable
  • DPP-wrapped

9. AIPDS (Deployment Specification)

Defines:

  • ≥1 DeploymentEnvironment
  • Runtime configuration
  • Optional training orchestration
  • Promotion gates
  • Resource profiles

AIPDS is:

  • Internal
  • Execution-oriented
  • Not exposed publicly

10. DPP (Digital Product Passport)

DPP:

  • Wraps AIPROD publish projection
  • Binds integrity hash
  • Records issuer identity
  • Enforces EffectiveRiskTier presence
  • Allows Inline or ByRef evidence
  • Provides revocation/supersession chain

DPP is product-centric.


11. Immutability Doctrine

For each AI Product version:

  • One AIPROD
  • One AIPDS
  • One EffectiveRiskTier
  • One DPP issuance state

Material changes require:

  • New artifact
  • Provenance linkage (prov:wasDerivedFrom)
  • New DPP issuance

No in-place mutation.


12. Architectural Positioning

AIPS is:

A computable AI regulatory operating system.

It is not:

  • A documentation template
  • A policy checklist
  • A compliance PDF

It is:

  • Ontology-anchored
  • SHACL-enforced
  • Version-immutable
  • Trust-bindable
  • Machine-enforceable