The previous post covered privacy, governance and regulation. The frameworks, information sources and practice ranges this series has drawn on are scattered across all of it, so here they are in one list. I hope it is useful.
AI security frameworks and information to look things up in#
- OWASP GenAI Red Teaming Guide: a testing methodology for generative AI. Follow it and go digging.
- OWASP AISVS: a verification checklist for AI systems, listing what to verify across twelve categories. Use it as the skeleton when you lay out test items.
- MITRE ATLAS: the AI version of ATT&CK, organizing attacks against AI into a matrix. Good for getting a feel for AI red team tactics.
- Google SAIF: Google’s Secure AI Framework, written from the builder’s side on how to wire security into an AI system.
- NIST AI RMF: AI governance in one place, answering who is responsible at which step and what evidence gets kept.
- Simon Willison’s blog: a deep archive on AI security in practice, with a steady stream of writing on what is new.
Free AI practice ranges to try your hand at#
- Break The Prompt: the range I worked through across two posts starting from AI Dark Arts (11).
- Lakera Agent Breaker: where the three MCP tool poisoning levels in AI Dark Arts (19) live. The same company also runs public ranges like Gandalf.
- PortSwigger’s Web LLM attacks: PortSwigger, which anyone doing web penetration testing knows well, has LLM labs to practise on too.
- LLMVault: a range that runs locally with no API key needed.
A short wrap-up#
I hope what has come out of this is useful to you. The next post is the last in the series, looking back over the thirty days and saying a few words to anyone walking the same road.