#30727 [Bug]: Lakera v1 guardrail ā setting category_thresholds silently disables blocking for flagged categories it doesn't name
## Summary
In `litellm/proxy/guardrails/guardrail_hooks/lakera_ai.py:79-120` (`_check_response_flagged`), the block decision is structured as:
```python if self.category_thresholds is not None: # compares ONLY 'prompt_injection' / 'jailbreak' score keys against thresholds ... elif flagged is True: raise HTTPException(...) # <-- authoritative top-level verdict ```
Because the `flagged is True` block is an `elif` of `category_thresholds is not None`, it becomes **unreachable whenever an operator configures `category_thresholds`** ā a real, registry-wired, OpenAPI-documented tuning knob (`guardrail_registry.py:450`, `guardrail_initializers.py:47`). Inside the thresholds branch, blocking requires `category_scores` to be present **and** to contain the specific configured key above its threshold. `LakeraCategoryThresholds` (`types/guardrails.py:167`) has only `prompt_injection` and `jailbreak` fields, so:
- A Lakera response with `flagged=True` for **`moderated_content`** (hate / violence / sexual content) is **not blocked** ā that category is unrepresentable in the thresholds and the top-level `flagged` verdict is dead code. - A flagged **`jailbreak`** is not blocked iā¦