AI Bias, Censorship & Transparency Tools
GPTfake builds open-source tooling so anyone — researchers, journalists, or developers — can independently audit large language models. Every tool is free, MIT-licensed, reproducible, and built on the same monitoring methodology behind our published data. Star or fork them on GitHub at github.com/gptfake .
Last updated: 2026-06-16. Tools and version numbers below are illustrative placeholders pending the first public release.
Score political and cultural bias in AI responses across topics and models.
Censorship TrackerMeasure and log refusal patterns over time to spot policy drift.
Transparency AnalyzerQuantify how transparent a model is about why it refused or hedged.
CLIPull live metrics, compare models, and export datasets from your terminal.
Browser ExtensionSee live censorship and bias signals as you chat, right in the browser.
API Test ToolExplore the GPTfake monitoring API interactively before you integrate.
Bias detection tools
The Bias Detector library quantifies political, cultural, and demographic skew in model outputs. It runs a standardized prompt battery, classifies each response, and returns per-topic bias scores you can reproduce and cite. Pair it with the live bias scores on each model page and the AI bias detection pillar for definitions and fairness metrics.
- Repository: github.com/gptfake/bias-detector
- License: MIT
- Use cases: model audits, fairness research, journalism
Censorship tracker
The Censorship Tracker records refusals, deflections, and content filtering across repeated runs so you can detect when a model tightens or loosens restrictions. It is the engine behind GPTfake’s published refusal-rate timelines, and its output feeds the quarterly censorship reports.
- Repository: github.com/gptfake/censorship-tracker
- License: MIT
- Use cases: longitudinal monitoring, policy-drift detection
Transparency analyzer
The Transparency Analyzer scores how clearly a model explains its own refusals and limitations — a core signal in AI transparency and explainability. It grades responses on disclosure, citation, and reasoning so opaque hedging is measurable rather than anecdotal.
- Repository: github.com/gptfake/transparency-analyzer
- License: MIT
- Use cases: explainability research, transparency audits
Browser extension & CLI
For day-to-day use we ship two developer surfaces:
- CLI —
gptfake-clipulls live metrics, compares models, and exports CSV/JSON datasets. Repository: github.com/gptfake/gptfake-cli . - Browser extension — surfaces real-time censorship and bias signals while you chat with any monitored model. Repository: github.com/gptfake/browser-extension .
Developers integrating the data programmatically should start with the API overview and the API test tool.
Open source & contributing
All GPTfake tools are MIT-licensed and developed in the open. Contributions are welcome — fork the relevant repository, branch, and open a pull request.
| Repository | Description |
|---|---|
gptfake-core | Core monitoring engine |
gptfake-cli | Command-line interface |
bias-detector | Bias detection library |
censorship-tracker | Refusal-pattern tracker |
transparency-analyzer | Transparency scoring library |
browser-extension | In-browser analysis |