About AIbins

AIbins is a curated, searchable reference of offensive techniques against AI/LLM agents - modeled on GTFOBins. A "bin" here is a technique: an attack primitive against an agent or LLM application, not a Unix binary. Each entry documents how to recognise it, the method, template payloads, why it works, its impact, and the fix - cross-referenced to the OWASP Top 10 for LLM Applications, the OWASP Top 10 for Agentic Applications, the OWASP Agentic AI Threats taxonomy, MITRE ATLAS, and CWE.

Ethics & scope

Educational, authorized, and defensive use only. Every payload is a template - adapt targets to systems you are authorized to test, and never fire blind at out-of-scope infrastructure. Entries describe technique classes and their defenses so red teams can test and blue teams can fix. AIbins is a static reference; it does not attack anything, and it does not ship operational exploits aimed at real third parties.

Licensing

Code is licensed MIT. Catalog content is licensed CC BY-SA 4.0 - attributable and share-alike.

Contributing

Add a technique by copying data/techniques/_TEMPLATE.yaml, filling the required fields, and keeping payloads benign and defensive-framed. Verify any MITRE ATLAS ids against the live matrix before asserting them - the build rejects ids not present in meta.yaml. Run npm run validate, then open a PR (one technique per PR). See CONTRIBUTING.md.

Credits

Seed content ported from an AI Agent Security CTF cheat-sheet and enriched with publicly documented technique classes. Framework definitions belong to their respective projects (OWASP GenAI Security Project, MITRE ATLAS, MITRE CWE).