AIPCH07 — Trustworthy
“Trust in the AI Product is evidence-backed, current, machine-consumable, and appropriate to its intelligence mechanism and risk.”
What AIPCH07 is really asserting
AIPCH07 is broader than model performance or drift monitoring.
A trustworthy Product version exposes enough current evidence to establish that its qualification, dependencies, governance controls, assurance state, and mechanism-appropriate behavior remain credible.
Trust currentness
AIPS distinguishes at least:
- AI qualification currentness;
- dependency currentness;
- governance currentness;
- assurance currentness.
Behavior-quality evidence is mechanism-specific. Examples may include model performance/drift, symbolic rule consistency, reasoning quality, agent completion/safety, or Physical AI control/safety evidence.
Model drift is one possible trust signal; it is not a universal definition of trust.
Positive criteria
AIPCH07 is Met when:
- a valid AIPS-DPP or equivalent governed trust record exists;
- qualification and dependency evidence are current;
- effective risk and Governance Obligation Profile evidence are current;
- assurance evidence is current and appropriate to the Product constitution;
- material incidents, restrictions, or invalidations are visible rather than hidden;
- trust state can be consumed by automated governance/discovery processes.
Not Met
Examples include:
- stale or invalid AQE represented as trustworthy;
- missing dependency provenance;
- trust expressed only as a producer claim;
- model metrics used as the sole trust evidence for an agentic/symbolic/Physical AI Product;
- mandatory assurance or governance evidence is absent.
Evidence signals
- AIPS-DPP;
- AQE and dependency paths;
- Governance Obligation Profile;
- assurance/evaluation evidence;
- observability and incident state;
- Marketplace lifecycle/restriction state.
Agent decision rule
Met when trust can be independently established from current evidence appropriate to the Product’s actual constitution, risk, agency, embodiment and effects.