Evidence boundary
These are design scenarios, not verified production deployments. VCP can carry and verify policy material, while each application remains responsible for enforcement, privacy, safety, legal compliance, and outcome evaluation.
Learning objectives
By the end of this module, you will be able to:
- Recognize the kinds of problem VCP is designed to help with
- Map VCP capabilities to specific organizational needs
- Separate what VCP provides from what each application must still build and prove
8.1 — Healthcare: consistent values across hospital systems
A hospital network could sign one VCP bundle of declared values, such as respect for patient autonomy, and distribute that same bundle to its clinical support, patient communication, and administrative integrations. Each integration would still need clinical validation and controls that demonstrate how verified values affect model behavior.
When the hospital updates its ethics guidance, it could issue a new bundle and revoke the old one. Audit records would contribute traceable evidence; they would not prove clinical quality or legal compliance by themselves.
VCP components used: bundles, verification, revocation, and privacy-aware audit records. As Module 4 notes, hashing reduces exposure but does not remove your obligations.
8.2 — Education: age-appropriate, culturally sensitive AI tutoring
A global education platform could use VCP to carry a student's age bracket, cultural setting, and current situation as inputs to its policy. A responsible implementation would still require child-safety review, consent controls, data minimization (collecting only what the tutor needs), and direct outcome testing. The protocol pattern could include:
- A CSM-1 persona chosen by age bracket, such as Nanny at adherence level 5
(
N5, child safety) for younger students and Godparent at adherence level 3 (G3, ethical guidance) for older ones - Cultural context that adjusts communication style via the situational tier
- Personal context that responds to what the student declares, such as a
cognitive state of
overloadedor an emotional tone offrustrated - Audit records that let the platform show parents which verified constitution each session used (the records hold no tutoring transcripts)
VCP components used: CSM-1 personas, context adaptation, personal state dimensions (the R-line), and audit records.
8.3 — Enterprise: organizational values at scale
A multinational company could use a shared baseline bundle, with narrower department-specific bundles at higher composition layers. The company's authorization rules would decide who may apply each bundle, and its composition policy would decide how the bundles combine. The pattern could include:
- A corporate baseline constitution shared by all departments
- Department-specific constitutions composed on top
- Role-based rules, set by the company, for who may apply each department's constitution
- Central governance with decentralized application
VCP components used: multi-constitution composition and attestation. The governance workflow itself stays with the company.
8.4 — AI safety research: reproducible alignment experiments
Researchers studying alignment can use VCP bundles as one controlled input to an experiment. Reproducibility still requires pinned models, prompts, sampling parameters, tools, datasets, environment, and analysis. VCP can contribute:
- Exact, versioned, hash-verified constitutional specifications
- Comparison experiments: same model, different constitutions, measurable differences
- Audit records of the bundle verified for each run
- A shared bundle input for carefully controlled cross-model comparisons
VCP components used: bundles (versioning and hashing), verification (which confirms each run used identical bundle bytes), and audit records.
8.5 — Content platforms: user-directed values
A social platform could let users choose or write their own preference bundles within a separately enforced platform baseline. The pattern could include:
- User-chosen or user-authored constitutions composed with the platform's non-negotiable baseline
- Decision records that identify the verified bundle and application outcome
- User-facing explanations derived from the actual decision path
- Attestation evidence that independent policy and runtime checks can evaluate
VCP components used: multi-constitution composition, audit records, and attestation. The user-facing explanations are the platform's to build.
VCP works the same way in every one of these domains. It lets declared values travel between systems and be verified at each integration boundary. Showing that the resulting behavior is principled, safe, effective, or compliant takes application controls and independent evidence.