AI alignment / norm elicitationAI alignmentParticipation documented
Collective Constitutional AI
San Francisco, United States
Anthropic and CIP explicitly describe a multistage public-input process with about 1,000 U.S. adults whose collectively drafted principles were then used to fine-tune and evaluate a language model.
democratic AIpublic alignmentcollective intelligenceproblem formulationgovernancemodel trainingevaluation
- Region
- North America
- Lead organization
- Anthropic; Collective Intelligence Project
- Organization type
- multi-stakeholder partnership
- Technology group
- generative language models
- Activity status
- completed
- Start year
- 2023
- Last updated
- Mar 25, 2026
- Participation mode
- public deliberation + constitutional drafting for model alignment
- Participants
- roughly representative U.S. adults; researchers; civic deliberation partners
- Methods
- Polis deliberation; public input; constitutional drafting; model fine-tuning; evaluation
- AI lifecycle stages
- problem formulation; governance; model training; evaluation
- Evidence summary
- Anthropic and CIP explicitly describe a multistage public-input process with about 1,000 U.S. adults whose collectively drafted principles were then used to fine-tune and evaluate a language model.
- Atlas assessment
- Cautious · High confidence · evidence grade A
- Uncertainty
- Public input directly shaped training principles, but the process was bounded to a U.S. sample and ultimate governance remained within a commercial lab.
- Participation group
- Deliberative Co Design
- Technology description
- generative language models
- Funding
- Not documented
- Region of activity
- national
- Verification status
- Live Verified
Community input
Corrections, annotations, and private disclosure requests.
Collective Constitutional AI has 0 published community items.
Published corrections
Loading moderation history...
Published annotations
Loading annotations...
Submit a correction
Flag a metadata problem or submit a dispute for Collective Constitutional AI.
Submit an annotation
Add context without editing the record itself.
Request restricted disclosure
These requests are private and reviewed for later releases.
Related records
AI alignment / norm elicitationNorth America
STELAUnited States (multiple communities), United States
Open recordAlgorithmic governance / public servicesNorth America
AI Failure Cards for Homeless ServicesPittsburgh, United States
Open recordParticipatory auditing / LLM safetyNorth America
AI Red Teaming ChallengeLas Vegas, United States
Open record