AI security
How to review, threat-model and defend applications built on large language models: assistants, RAG systems, copilots with tools, and agents. Grounded in the OWASP GenAI lists, the NIST AI RMF and CISA/NCSC guidance, with every claim cited to the page it came from.
Guides
Prompt Injection Explained Through a RAG Assistant Scenario
What indirect prompt injection is, how a poisoned document turns a RAG assistant against its users, three real 2025–26 cases, and eight controls with where each one fails.
Read the guideComing next: Threat Modeling an LLM Application Step by Step · Securing AI Agents: Tools, Permissions and Human Approval · AI Security Governance: What Leaders Should Measure.
Free download: AI Application Security Review Checklist
28 checks, 15 of them marked Deterministic because they hold even when the model is fooled. PDF plus an editable spreadsheet with a self-totalling summary.
Get the checklistGoing further
The checklist tells you where to look. The AI Security Assessment Toolkit adds the threat-model worksheet, risk register, control mapping, evaluation plan and report template that turn the answers into a review your leadership can act on.
Independent and vendor-free: see our disclosures.
