Governance & Evaluation
Model Card Template
Purpose
Standardizes documentation published alongside each model release.
Scope
All publicly and enterprise-released models.
Definitions
Model Card.
Policy Statements
Each Model Card includes: Model Name, Family, Version, Release Date, Training Methodology, Intended Uses, explicitly enumerated Out-of-Scope Uses (per the Acceptable Use Policy's foreseeable-use exclusion requirement), Capabilities, Limitations, Known Risks, Failure Modes, Benchmark Results (citing specific Benchmark Methodology Policy versions), Safety Evaluation Results (including CBRN-gate outcome where applicable), Security Considerations, Privacy Considerations (including training-data opt-out status), Environmental Impact (compute/energy footprint, per Doc. 39), and a Changelog.
Responsibilities
Research Evaluation team drafts; AI Safety Research Lead and Technical Writing review before publication.
Compliance Requirements
No public launch without a published Model Card.
Enforcement
Missing/inaccurate Model Cards block public release.
Reporting Process
Exceptions Process
Enterprise-only/restricted releases may use abbreviated cards with Governance approval.
Review Process
Updated with every model version release.