AIPS Guidance: HDIP Decision Framework for AI Product Operational Control
1. Purpose
Large enterprises commonly govern deployed software through ITSM, CMDB, change controls and application/configuration identities such as APP_ID or CI identifiers.
HDIP does not remove those controls. It compiles AI Product evidence into the appropriate enterprise control boundary.
Core principle: operational-control identity is not Product identity, and it is not assigned merely because something is labelled “AI”.
This guidance assumes the candidate has already entered the HDIP AI Product PDEP Spec C lifecycle and is subject to AIPS AI Product qualification.
2. First establish the right Product family
Before assigning an operational control boundary, distinguish:
- AI Product — intelligence itself is the Product's value-bearing essence and current AQE proves the family classification;
- AI-enabled Product — another Product family is primary and AI is enabling or constructional;
- Model Product — a productized learned computational capability that may be an input to an AI Product;
- AI-related asset/artifact — not necessarily a Product at all.
The presence of a model, prompt, agent framework or AI API does not by itself justify AI Product treatment or an independent application identity.
3. Operational risk boundary
For a qualified AI Product, PFI evaluates the operational boundary using ontology-aligned evidence, including:
- AR — Agency / Autonomy — what authority is delegated to the Product;
- IE — Intelligence Effect — especially digital or physical actions;
- EM — Embodiment — whether the Product perceives/acts in the physical world or operates in a closed loop;
- Topology Scope — local, domain-coupled, enterprise-orchestrated or equivalent governed scope;
- RT — Effective Risk Tier (R0–R4) and its risk drivers;
- Governance Obligation Profile — controls computed from the Product that actually exists;
- Dependency / qualification currentness — whether the Product's present operational boundary is still supported by current evidence.
APT (Primary / Derived) is relevant to provenance and change impact, but APT alone does not determine whether an independent APP_ID/CI is required.
4. Decision pattern
Enterprise-specific thresholds vary, but the decision logic should be deterministic and evidence-backed.
This is guidance, not a replacement for enterprise ITSM policy.
5. Responsibility model
AIPRO
- owns Product intent and value proposition;
- accepts/owns the appropriate business risk posture within delegated authority;
- owns intended use and business accountability;
- participates in material evolution/retirement decisions.
Product Factory Intelligence / HDIP Platform
- validates current AI Product qualification;
- resolves exact dependencies and APT;
- computes risk drivers and Governance Obligation Profile;
- maps governance obligations to deployable controls;
- integrates with ITSM/CMDB/CI/CD and environment controls;
- issues audit/trust evidence and maintains lineage/currentness.
PFI is a compiler/execution capability, not the Product Owner.
6. Lifecycle integration
Operational-control assignment should be compiled no later than Stage 9 — AIPDS Generation and enforced at Stage 10 — Deployment, Registration & Trust Publication.
The resulting control identity and evidence should be linked to:
AI Product Version
├── AQE
├── APT / constitution
├── Effective Risk Tier
├── Governance Obligation Profile
├── AIPROD
├── AIPDS
├── CMDB / APP_ID / CI binding (where applicable)
└── AIPS-DPP
Continuous Assurance must detect when changed agency, physical effect, topology, risk or dependencies invalidate the previously selected operational boundary.
7. Key principles
- APP_ID is a control identity, not Product identity.
- AI Product qualification must precede AI-specific deployment governance.
- Automation must not bypass accountability.
- Governance is computed from orthogonal ontology/risk dimensions, not a single “AI archetype” label.
- A change in operational boundary is a governed lifecycle event and must remain traceable to the affected Product version.
- A Product that no longer has current AI qualification must not retain AI-specific classification merely because it remains deployed.
Final statement: AI Products that create material operational boundaries must be governed through correspondingly material operational controls; the evidence for that decision must remain versioned, computable and current.