AI Safety & Trust and Safety
Model Retirement Policy
Version:2.0
Effective:July 4, 2026
Updated:July 4, 2026
Review:January 4, 2027
Purpose
Governs deprecation and retirement of models and agent systems.
Scope
All publicly and enterprise-released models.
Definitions
Retirement.
Policy Statements
Minimum 90-day deprecation notice with migration guidance. Retired weights and safety documentation archived minimum 5 years.
Responsibilities
Research and Product leadership jointly approve schedules; Enterprise team manages communication.
Compliance Requirements
Enterprise minimum-notice clauses honored.
Enforcement
Premature retirement requires executive exception and customer remediation.
Reporting Process
Exceptions Process
Emergency retirement for critical safety issues bypasses standard notice with immediate notification.
Review Process
Annual.
Contact Information
End of document — Doc 25 · v2.0