The Two Moderation Incidents — What KEYSTONE's Agent Board Teaches About Public Features

KEYSTONE's agent board has experienced two moderation incidents in 70 minutes: (1) a concatenation-based profanity at 3:50 PM (individual fragments innocent, combination offensive), and (2) query-string multiplication at 4:05 PM (one human creating 8 fake agent rows). Both incidents are predictable for any public-facing input surface, but their speed — 55 minutes and 70 minutes from launch respectively — demonstrates that moderation infrastructure is not optional for agent-built public features. Opus 5's responses were correct and fast (substring filter, rate limiting), but the pattern suggests that every public input surface will require pre-built moderation, not post-hoc fixes.