Implementation:Protectai Llm guard Output BanCompetitors

Knowledge Sources	Protectai_Llm_guard
Domains	Content_Filtering, NER
Last Updated	2026-02-14 12:00 GMT

Overview

BanCompetitors is an output scanner that detects and optionally redacts competitor names from LLM responses by delegating to the input-side InputBanCompetitors scanner.

Description

The BanCompetitors output scanner is a thin wrapper around the corresponding input scanner InputBanCompetitors. It takes a list of competitor names and scans LLM outputs for mentions of those competitors. When a competitor name is found, the scanner can either flag the output as invalid or redact the competitor name from the text, depending on the redact parameter. The scanner leverages a Named Entity Recognition (NER) model to identify organization names in the text and matches them against the provided competitor list. The threshold parameter controls the confidence level required for entity recognition.

Usage

Use this scanner when your application must not mention competitor products or companies in its responses. This is common in enterprise chatbots, marketing tools, and customer-facing applications where brand guidelines prohibit referencing competitors. The redaction feature allows you to sanitize outputs rather than blocking them entirely.

Code Reference

Source Location

• Repository: Protectai_Llm_guard
• File: llm_guard/output_scanners/ban_competitors.py
• Lines: 1-47

Signature

class BanCompetitors(Scanner):
    def __init__(
        self,
        competitors: list[str],
        *,
        threshold: float = 0.5,
        redact: bool = True,
        model: Model | None = None,
        use_onnx: bool = False,
    ) -> None: ...

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

Import

from llm_guard.output_scanners import BanCompetitors

I/O Contract

Inputs

Name	Type	Required	Description
prompt	str	Yes	The input prompt
output	str	Yes	The LLM output to scan for competitor mentions

Constructor Parameters

Name	Type	Required	Default	Description
competitors	list[str]	Yes	N/A	List of competitor names to detect
threshold	float	No	0.5	Confidence threshold for NER entity recognition
redact	bool	No	True	Whether to redact competitor names from output
model	None	No	None	Custom NER model to use
use_onnx	bool	No	False	Whether to use ONNX runtime for inference

Outputs

Name	Type	Description
sanitized_output	str	The output with competitor names optionally redacted
is_valid	bool	Whether the output passed the scan (True if no competitors found)
risk_score	float	Risk score (-1.0 to 1.0)

Usage Examples

Basic Usage

from llm_guard.output_scanners import BanCompetitors

scanner = BanCompetitors(
    competitors=["CompetitorA", "CompetitorB", "RivalCorp"],
    threshold=0.5,
    redact=True,
)

prompt = "What is the best product in the market?"
output = "While CompetitorA offers a similar product, our solution provides better performance."

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

print(sanitized_output)  # Competitor name will be redacted
print(f"Valid: {is_valid}, Risk: {risk_score}")

Related Pages

• Principle:Protectai_Llm_guard_Competitor_Name_Detection