Safety & Red Teaming Cheat Sheets

Bias and Fairness Metrics cheat sheet

When groups have different base rates, you choose which fairness to satisfy and write down the trade-off.

#Safety & Red TeamingAnimated

→ Topic: Bias and Fairness Metrics

Class Imbalance cheat sheet

A model trained on skewed data can look accurate while missing the rare cases; change the metric before you change the data.

#Safety & Red TeamingAnimated

→ Topic: Class Imbalance

Data Drift cheat sheet

A drift alert says the data moved, not that accuracy fell, so verify before you retrain.

#Safety & Red TeamingAnimated

→ Topic: Data Drift

Data Leakage cheat sheet

A transformation fit before the split grades the model on answers it has already seen.

#Safety & Red TeamingAnimated

→ Topic: Data Leakage

Data Poisoning cheat sheet

Training-time attacks need retraining to fix; RAG and agent memory stay open after deployment.

#Safety & Red TeamingAnimated

→ Topic: Data Poisoning

Data Versioning cheat sheet

A pointer in Git plus bytes in cheap storage rebuild any past dataset, but the hash cannot tell good data from bad.

#Safety & Red TeamingAnimated

→ Topic: Data Versioning

Dataset Bias cheat sheet

Collecting more data from the same source only enlarges the skew; a representative sample is what fixes it.

#Safety & Red TeamingAnimated

→ Topic: Dataset Bias

Hallucination cheat sheet

A language model predicts what sounds true; grounding and claim checks shrink the gap but never close it.

#Safety & Red TeamingAnimated

→ Topic: Hallucination

Red Teaming for AI cheat sheet

Every attack that lands becomes a guardrail, and the cycle starts again after each model or prompt change.

#Safety & Red TeamingAnimated

→ Topic: Red Teaming for AI

Toxicity and Safety Evaluation cheat sheet

Benchmarks only see known attacks, so a safety check repeats after every model or prompt change.

#Safety & Red TeamingAnimated

→ Topic: Toxicity and Safety Evaluation