NIST AI RMF and the OWASP LLM Top 10
Short answer: The NIST AI Risk Management Framework (AI RMF 1.0) is a free, voluntary guide from the US National Institute of Standards and Technology for managing the risks of AI. It is built around four functions: Govern, Map, Measure and Manage. The OWASP Top 10 for LLM Applications lists the most common security risks in apps built on large language models, starting with prompt injection.
What is the NIST AI RMF?
NIST published AI RMF 1.0 in January 2023, and in July 2024 added a Generative AI Profile (NIST AI 600-1) with guidance specific to generative AI. The framework is not a law and there is no certificate. It is a shared vocabulary and a set of practices that help organisations build AI that is valid and reliable, safe, secure and resilient, accountable and transparent, explainable, privacy-enhanced and fair.
What are the four functions of the NIST AI RMF?
| Function | What it means in practice |
|---|---|
| Govern | Set policies, roles, accountability and a culture of managing AI risk. Govern applies across the other three. |
| Map | Understand the context: what each AI system is for, who it affects and what could go wrong. |
| Measure | Test and track the risks you mapped, for example accuracy, bias, robustness and security. |
| Manage | Prioritise and act on the risks, respond to incidents and keep monitoring after launch. |
What is the OWASP Top 10 for LLM Applications?
OWASP, the open community behind the well-known web application Top 10, also publishes a list for applications built on large language models. The 2025 edition is:
- Prompt injection: crafted input makes the model ignore its instructions.
- Sensitive information disclosure: the model reveals private data or secrets.
- Supply chain: risky third-party models, datasets or plugins.
- Data and model poisoning: tampered training or fine-tuning data.
- Improper output handling: model output is trusted and passed to code, browsers or databases without checks.
- Excessive agency: the model can take actions with more permissions than it needs.
- System prompt leakage: hidden instructions or secrets in the system prompt are exposed.
- Vector and embedding weaknesses: problems in retrieval (RAG) pipelines, such as data leaking between users.
- Misinformation: confident but wrong answers that people rely on.
- Unbounded consumption: uncontrolled use that runs up cost or takes the service down.
How do you get started with AI risk management?
- Make a list of AI systems you build and use, including tools staff use on their own.
- Name an owner for each and agree an AI use policy.
- For each system, write down its purpose, its users, the data it touches and the worst realistic failure.
- Test for the OWASP LLM risks above, especially prompt injection and improper output handling.
- Limit what AI features can do: least privilege for tools and APIs, human approval for risky actions.
- Monitor in production and have a way for users to report problems.
If you want a certificate to show customers, the next step is ISO/IEC 42001.
How ExpertComply helps
ExpertComply GRC includes a NIST AI RMF 1.0 checklist and an ISO/IEC 42001 checklist, with AI governance and AI risk templates. ExpertComply VAPT scans your code for LLM risks such as prompt injection and unsafe handling of model output, and explains each fix with a prompt you can paste into your AI coding tool. Read more on AI security.
See plans Ask for a walkthrough