Free checklist · AI security
The AI security checklist for LLM applications.
Twenty-four checks for chatbots, copilots and AI agents before they meet real users and real data, organised around the OWASP Top 10 for LLM Applications.- 01The ten LLM risks, each with what goes wrong in plain words.
- 02Two practical checks per risk that your team can test.
- 03Governance checks for regulated companies.
- 04A simple way to see which risks are still open.
- LLM apps
- AI agents
- Copilots
- RAG search
Get the guide
Free, and on screen in a few seconds.
The ten LLM security risks.
- LLM01 Prompt injectionInstructions hidden in user input or documents change what the model does.
- LLM02 Sensitive information disclosureThe model reveals personal or confidential data.
- LLM03 Supply chainThird-party models, plugins or packages carry hidden risks.
- LLM04 Data and model poisoningTampered training or retrieval data changes model behaviour.
- LLM05 Improper output handlingModel output flows into other systems without checks.
- LLM06 Excessive agencyAn AI agent has more permissions or autonomy than it needs.
- LLM07 System prompt leakageSecrets or rules written into the prompt get exposed.
- LLM08 Vector and embedding weaknessesRetrieval returns documents the user should not see.
- LLM09 MisinformationConfident but wrong answers are trusted.
- LLM10 Unbounded consumptionUncontrolled usage drives up cost or takes the service down.