AI moderation is good at scale and bad at nuance. The AI Moderation Policy defines what AI moderation can do alone (flag, queue, sort, prioritize), what it can never do alone (suspend, ban, remove content from accused individuals, publish identifying information, escalate to law enforcement), and how AI moderation decisions are audited.
It applies to the Map.ca moderation team, the AI team that builds moderation tooling, and any vendor providing moderation AI.
Requirements
- Restrict AI moderation to flag, queue, sort, and prioritize actions when acting alone.
- Require human review before any irreversible moderation action.
Prohibitions
- Do not let AI alone suspend, ban, or remove content based on identity or accusation.
- Do not deploy AI moderation that lacks an audit trail.