Constitutionally Aligned Model

Model assetModel AdaptationModelsarc:ConstitutionallyAlignedModel

A language model checkpoint whose weights were trained by critique-revision supervised fine-tuning and AI-feedback preference optimization to adhere to a constitution.

Responsibility. Generates responses that embody constitutional principles by default.

Also known as: CAI-trained model, RL-CAI policy, Aligned policy model

Direct neighbourhood (hover for relationship types)

Relationships

deployed on structural

LLM Inference Service Ch9.5

is evaluated by assurance

is monitored by assurance

Alignment Drift Monitor Ch9.5

is trained by lifecycle

Fine-Tuning Pipeline abstract Ch9.5
RLHF Policy Optimizer Ch9.5

Design guidance

MUST be continuously monitored and periodically retrained; alignment degrades through distribution shift, fine-tuning drift, adversarial attacks, interpretation drift and catastrophic forgetting.
SHOULD include explicit alignment maintenance whenever domain fine-tuning is applied.

Classification

Patterns: Training-time alignmentHelpful, harmless, honest (HHH)
Quality attributes: Safety (ISO/IEC 25010 | NIST AI RMF: safe)Reliability (ISO/IEC 25010 | NIST AI RMF: valid and reliable)Functional suitability: correctness and validity (ISO/IEC 25010 | NIST AI RMF: valid)
Risks mitigated: Prompt-only safety overridden by jailbreaks or injectionHelpfulness-harmlessness trade-off

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.