Constitution

Data artifactGovernance & ComplianceSafety, Security & GovernanceVariation point (abstract)arc:Constitution

A versioned, human-readable set of explicit natural-language principles (e.g., helpful, harmless, honest; domain rules) that guides model training and runtime behaviour and can be publicly inspected and debated.

Responsibility. States the explicit values an AI system must adhere to.

Also known as: Constitutional principles, Principle set, Explicit value specification, Value documentation, Global value specification, Value specification

Direct neighbourhood (hover for relationship types)

Variants

Variant	When to choose
Regional Constitution	Choose for globally deployed systems whose jurisdictions differ in law and cultural norms (e.g., political speech, religious expression), accepting higher design, evaluation and maintenance cost.
Unified Constitution	Choose when one consistent value set can serve all users, explicitly acknowledging limited scope (particular value commitments) rather than claiming universal validity.
User-Configurable Constitution	Choose when users legitimately differ in how they weight values such as privacy versus convenience, and customization can be bounded by non-negotiable principles.

Relationships

configures structural

is read by dependency

receives data from dynamic

constrains control

Agent Controller abstract Ch9.5
Cross-System Record Conflict Resolver Ch9.5

is audited by assurance

Design guidance

SHOULD draw principles from multiple sources rather than a single organization's values, and disclose whose values are represented and which are deliberately not prioritized.
SHOULD extend general principles with domain-specific principles (e.g., clinical referral, fiduciary disclosure, fair lending) for regulated deployments.
MUST NOT be treated as a guarantee of behaviour: principles make intended values inspectable, not the learned implementation auditable.
SHOULD be reviewed on a schedule and updated as deployment contexts evolve and new risks emerge.
SHOULD be combined with defense-in-depth safeguards (moderation, guardrails, human review, monitoring) because principle ambiguity, incomplete coverage and value conflicts are irreducible.
MUST record whose values are represented, how stakeholders were engaged and cultural considerations for diverse jurisdictions (EU AI Act value documentation).
SHOULD be identified with affected communities, not assumed by technical teams alone.
SHOULD be versioned and updated through a documented process as societal norms evolve.

Classification

Patterns: Constitutional AIExplicit value encodingMulti-source principle derivationDomain-specific principle extensionCore vs. culturally-adapted principlesTop-down value alignment
Quality attributes: Transparency and accountability (NIST AI RMF: accountable and transparent)Maintainability (ISO/IEC 25010)Reliability (ISO/IEC 25010 | NIST AI RMF: valid and reliable)
Risks mitigated: Opaque implicit values learned from preference dataUntraceable alignment failuresUnauthorized policy exceptionsHarmful, biased or deceptive outputsUnstated designer assumptions about values
Frameworks & regulations: EU AI Act: transparency requirementsUN Declaration of Human Rights (principle source)Equal Credit Opportunity ActFair Housing ActGDPRSecurities regulations / fiduciary dutyEU AI Act (high-risk requirements)IEEE Ethically Aligned Design

Sources

Ch9.5: T. Nguyen, "Constitutional AI," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 9.5. ISBN: 9798244538229.
Ch9.6: T. Nguyen, "Value Alignment Frameworks," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 9.6. ISBN: 9798244538229.
Ch10.3: T. Nguyen, "RLHF Methodology," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 10.3. ISBN: 9798244538229.