Constitutional AI Framework
Building transparent, accountable, and fair AI systems through structured governance principles and hierarchical rules
2
Constitutions Published
2
Countries Covered
5
Tier Framework
100%
Human Oversight
Sample Constitutions
Explore our growing library of AI governance frameworks
More Coming Soon
We are developing constitutions for healthcare AI, educational assessment systems, criminal justice algorithms, and social services automation. Each framework is designed to ensure AI serves human values while protecting fundamental rights.
We are developing constitutions for healthcare AI, educational assessment systems, criminal justice algorithms, and social services automation. Each framework is designed to ensure AI serves human values while protecting fundamental rights.
About Constitutional AI
A governance framework for responsible artificial intelligence
What is Constitutional AI?
Constitutional AI is an approach to AI governance that embeds ethical principles, legal requirements, and human values directly into the design, deployment, and operation of AI systems. Like a national constitution provides the foundational rules for government, a Constitutional AI framework provides the foundational rules for AI systems operating in specific domains.
Rather than treating AI ethics as an afterthought or relying solely on voluntary industry standards, Constitutional AI creates enforceable, hierarchical frameworks that ensure AI systems respect human rights, operate transparently, and remain accountable to the people they affect.
Why Constitutional AI Matters
As AI systems increasingly make decisions that affect people's lives—from credit approvals to hiring decisions to healthcare diagnoses—we need robust governance frameworks that ensure these systems serve human values and protect fundamental rights. Constitutional AI provides:
Transparency
Clear explanations of how AI systems work, what data they use, and how they make decisions
Accountability
Defined responsibility for AI decisions and clear pathways for remedy when things go wrong
Fairness
Protection against discrimination and bias, with continuous monitoring and correction
Human Agency
AI assists but never replaces human judgment in consequential decisions
Privacy Protection
Data minimization and strong safeguards for personal information
Continuous Improvement
Living frameworks that evolve with evidence and community input
Core Principles
Foundational values that guide all Constitutional AI frameworks
1. Human Primacy
AI systems serve people, not the other way around. Humans maintain ultimate decision-making authority in consequential matters. AI recommendations are advisory; humans make final judgments. Every person affected by AI has the right to request human review and to challenge automated decisions.
2. Transparency & Explainability
People have the right to know when AI is being used, how it works, and why it made specific decisions affecting them. Explanations must be in plain language that ordinary people can understand, not technical jargon accessible only to engineers. The logic behind AI decisions must be intelligible to those affected.
3. Fairness & Non-Discrimination
AI systems must not discriminate based on protected characteristics such as race, gender, age, disability, religion, or other identity factors. Assessment accuracy must be equal across all demographic groups. Proxy variables that correlate with protected characteristics are prohibited. Disparate impacts trigger immediate investigation and correction.
4. Privacy & Data Protection
Collect and use only data necessary for legitimate purposes. Protect personal information with strong safeguards. No secondary uses without explicit consent. People have the right to access, correct, and delete their data. Data security is mandatory, not optional.
5. Accountability & Remedy
Organizations deploying AI are accountable for its decisions and impacts. When AI causes harm, clear pathways exist for understanding what happened, challenging the decision, and seeking remedy. Accountability cannot be outsourced to algorithms or vendors.
6. Inclusion & Accessibility
AI systems must be designed to serve diverse populations, including those historically excluded or marginalized. Accessibility accommodations are required, not optional. Cultural contexts are respected. Alternative documentation and assessment methods are available when formal processes create barriers.
7. Continuous Monitoring & Improvement
AI systems are continuously monitored for bias, errors, and unintended consequences. When problems are detected, they are corrected immediately. Frameworks evolve based on evidence, not static assumptions. Regular audits by independent third parties ensure compliance and identify areas for improvement.
Framework Structure
How Constitutional AI frameworks are organized
All Constitutional AI frameworks follow a hierarchical five-tier structure. Each tier builds on the previous one, moving from abstract principles to concrete operational rules.
Tier 1: Foundational Principles
Core values that guide all decisions. These are the "why" behind the framework—the fundamental commitments to human dignity, fairness, transparency, and accountability that cannot be compromised.
Tier 2: Operational Principles
How foundational principles are implemented in practice. These translate abstract values into concrete operational requirements, such as data minimization, explainability standards, and monitoring protocols.
Tier 3: Specific Rules
Detailed implementation rules for concrete scenarios. These provide specific guidance for particular AI systems, use cases, and decision points—the "how-to" of Constitutional AI.
Tier 4: Monitoring & Learning
Systems for tracking performance, detecting problems, and driving continuous improvement. Defines what metrics to track, how to report them, and how to use insights to make the system better over time.
Tier 5: Amendment & Governance
How the framework itself evolves. Defines review cycles, amendment processes, stakeholder participation, and governance structures to ensure the constitution remains living and responsive to new evidence.
Hierarchical Design
This five-tier structure ensures that all rules derive from clear principles, making the framework coherent and internally consistent. When conflicts arise, higher tiers take precedence. When new situations emerge, they can be addressed by applying the established principles through the hierarchy.
This five-tier structure ensures that all rules derive from clear principles, making the framework coherent and internally consistent. When conflicts arise, higher tiers take precedence. When new situations emerge, they can be addressed by applying the established principles through the hierarchy.
Key Components
In addition to the five-tier structure, each Constitutional AI framework includes:
Rights Declarations
Clear statements of the rights held by people affected by the AI system (borrowers, job seekers, patients, etc.)
Permitted & Prohibited Data
Explicit lists of what data can and cannot be used, preventing mission creep and protecting privacy
Escalation Triggers
Conditions that require immediate human review, ensuring AI doesn't operate unchecked in edge cases
Fairness Thresholds
Quantitative standards for acceptable performance disparities across demographic groups
Appeal Mechanisms
Clear processes for challenging AI decisions and obtaining human review
Transparency Requirements
Mandatory disclosures, explanations, and reporting to ensure public accountability