A field guide to making AI models misbehave

What it does This repository is a curated list of resources covering Offensive AI: adversarial machine learning attacks, generative AI abuse, and practical tooling for pentesters and researchers. It catalogs extraction, inversion, poisoning, and evasion attacks alongside defensive countermeasures, plus sections on malware, phishing, OSINT, and generative AI misuse across audio, image, video, and text modalities.

The interesting bit The README doesn’t just dump links—it taxonomizes. Each attack category (extraction, inversion, poisoning, evasion) gets its own conceptual diagram, limitations section, defensive actions, and annotated paper trail. It’s a textbook masquerading as a GitHub repo, and the generative AI section reads like a sober inventory of capabilities that sci-fi warned us about.

Key highlights

• Four core adversarial ML attack types with visual explainers: extraction, inversion, poisoning, evasion
• Generative AI misuse broken down by medium: audio deepfakes, image manipulation, video synthesis, text generation
• Practical tooling sections including ART and Cleverhans frameworks
• Each attack category paired with specific defensive actions (differential privacy, homomorphic encryption, ensemble methods)
• Extensive paper links with arXiv/ACM/IEEE sourcing, plus related GitHub repos with star counts

Caveats

• The README is heavy on emoji taxonomy; navigation requires patience
• Some sections (notably evasion and backdoors) were truncated in the provided source
• No code in the repo itself—this is purely a reference compilation

Verdict Security researchers, red teamers, and ML engineers building threat models should bookmark this. If you’re looking for runnable exploits, look elsewhere; this is the map, not the territory.