Why it matters

Built-in filters are first line of safety. Understanding shapes right configuration.

Advertisement

The architecture

Categories: harassment, hate speech, sexually explicit, dangerous.

Thresholds: BLOCK_NONE, LOW, MEDIUM (default), HIGH.

Gemini safety filtersCategories4 harm typesThresholdsNONE / LOW / MED / HIGHApplied toinput + outputDefault MEDIUM balances safety + false positives
Filter configuration.
Advertisement

How it works end to end

Filters applied to prompt + response.

Configuration per request or session.

Custom safety: combine with app-level guards.

False positives on legitimate content possible.