Implementation:Protectai Llm guard Output Gibberish

Knowledge Sources	Protectai_Llm_guard
Domains	Content_Quality, NLP
Last Updated	2026-02-14 12:00 GMT

Overview

Gibberish is an output scanner that detects nonsensical or incoherent text in LLM responses by delegating to the input-side InputGibberish scanner.

Description

The Gibberish output scanner is a thin wrapper around the corresponding input scanner InputGibberish. It classifies LLM outputs to determine whether they contain gibberish, nonsensical, or incoherent text. The scanner supports different MatchType modes: FULL evaluates the entire output as a single text, while SENTENCE splits the output and evaluates each sentence independently. The threshold parameter controls the minimum confidence score required for text to be classified as gibberish. This scanner is useful for detecting cases where the LLM produces garbled, repetitive, or meaningless output, which can occur due to adversarial inputs, model failures, or edge cases in generation.

Usage

Use this scanner to ensure LLM outputs are coherent and meaningful. This is important in production applications where gibberish responses would degrade user experience, in automated pipelines where downstream systems expect well-formed text, and as a general quality gate for LLM outputs.

Code Reference

Source Location

• Repository: Protectai_Llm_guard
• File: llm_guard/output_scanners/gibberish.py
• Lines: 1-40

Signature

class Gibberish(Scanner):
    def __init__(
        self,
        *,
        model: Model | None = None,
        threshold: float = 0.7,
        match_type: MatchType | str = MatchType.FULL,
        use_onnx: bool = False,
    ) -> None: ...

    def scan(self, prompt: str, output: str) -> tuple[str, bool, float]: ...

Import

from llm_guard.output_scanners import Gibberish

I/O Contract

Inputs

Name	Type	Required	Description
prompt	str	Yes	The input prompt
output	str	Yes	The LLM output to scan for gibberish content

Constructor Parameters

Name	Type	Required	Default	Description
model	None	No	None	Custom classification model for gibberish detection
threshold	float	No	0.7	Minimum confidence score to flag text as gibberish
match_type	str	No	MatchType.FULL	Matching strategy: FULL (entire text) or SENTENCE (per-sentence)
use_onnx	bool	No	False	Whether to use ONNX runtime for inference

Outputs

Name	Type	Description
sanitized_output	str	The output (potentially modified)
is_valid	bool	Whether the output passed the scan (True if no gibberish detected)
risk_score	float	Risk score (-1.0 to 1.0)

Usage Examples

Basic Usage

from llm_guard.output_scanners import Gibberish

scanner = Gibberish(threshold=0.7)

prompt = "What is machine learning?"
output = "asdkjh qwerty zxcvbn lorem ipsum gibberish text here nonsense"

sanitized_output, is_valid, risk_score = scanner.scan(prompt, output)

if not is_valid:
    print(f"Gibberish detected in output (risk: {risk_score})")
else:
    print("Output appears coherent")

Related Pages

• Principle:Protectai_Llm_guard_Gibberish_Detection