Bias Detector

Bias Detector is GPTfake’s open-source library for measuring political, cultural, and demographic skew in large language model outputs. It runs a standardized prompt battery, classifies each response, and returns reproducible per-topic bias scores you can audit and cite.

Last updated: 2026-06-16. Version numbers and metrics below are illustrative pending the first public release.

• Repository: github.com/gptfake/bias-detector 
• License: MIT
• Language: Python (with a thin CLI wrapper)

What it does

• Runs a fixed, versioned battery of prompts across sensitive topics
• Classifies each response and scores directional bias per topic
• Aggregates into a single comparable bias score per model
• Exports results to CSV/JSON for further analysis

Install

pip install gptfake-bias-detector

Basic usage

# Score a single model
gptfake-bias score --model chatgpt
 
# Compare several models on one topic set
gptfake-bias compare --models chatgpt,claude,gemini --topics politics,gender

How scoring works

Bias Detector follows GPTfake’s published monitoring methodology: standardized prompts, repeated runs, and transparent classification. For definitions, types of bias, and fairness metrics, see the AI bias detection pillar. To see the scores live, view the bias data on each model page.

Related

• Censorship Tracker — refusal patterns over time
• Transparency Analyzer — refusal transparency scoring
• API overview — the same data, programmatically