EthicAI — AI Ethics Auditor

Audit your ML models for bias, robustness & EU AI Act compliance — in 30 seconds, without leaving VS Code.

Bias detection · SHAP explainability · Robustness testing · Compliance reports — all running locally on your machine.

Why EthicAI?

The EU AI Act requires high-risk AI systems to prove they are fair, explainable, robust, and documented. Traditional audits cost €50k–200k and take months. Cloud tools send your test data off-machine — a GDPR risk.

EthicAI brings the audit to the developer. Open your model file, click once, and get a full ethics report in seconds — bias metrics, feature attributions, stress tests, and an EU AI Act checklist — with zero data leaving your machine.

Think SonarQube for code, but for the ethics of your ML models.

✨ Features

🚀 Quick Start

1. Install the extension

From the VS Code Marketplace, or:

code --install-extension ethicai.ethicai

2. Install the Python engine

pip install ethicai

3. Sign in

On first launch you'll see a Sign in with Google screen. Authenticate once — your token is stored securely in VS Code's SecretStorage.

4. Run an audit

Open a Python file or notebook, click the 🛡️ EthicAI icon (activity bar or editor title), pick your file, and watch the report appear.

📝 Your model file just needs four variables

EthicAI imports your script as a module and looks for these at the top level:

model              # any trained sklearn-compatible model
X_test             # test features (NumPy array or DataFrame)
y_test             # true labels (1-D array)
sensitive_features # optional — e.g. df["gender"], enables bias analysis

⚠️ Define them at module level (not inside if __name__ == "__main__":), and avoid blocking calls like plt.show() or input().

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
model = RandomForestClassifier().fit(X_train, y_train)
sensitive_features = df.loc[X_test.index, "gender"]

🤖 Supported Models

from ethicai import TorchModelWrapper, TFModelWrapper

model = TorchModelWrapper(torch_net)   # or TFModelWrapper(keras_model)

📋 EU AI Act Coverage

EthicAI maps each audit result to the EU AI Act (Regulation 2024/1689) — 12 checks across 9 articles:

🔒 Privacy-First

Your data never leaves your machine. EthicAI runs a local Python subprocess — no cloud, no API keys, no uploads. Training and test data stay on disk. This is what makes EthicAI usable where cloud auditing tools are legally off-limits.

⚙️ Commands & Settings

Commands (Command Palette → "EthicAI")

Settings

💡 If you use conda/venv, set ethicai.pythonPath to the absolute interpreter path (e.g. /opt/miniconda3/bin/python) so the extension finds the package regardless of how VS Code was launched.

📦 Requirements

• VS Code 1.85+
• Python 3.10+ with the ethicai package (pip install ethicai)

🐛 Troubleshooting