We publish this page because NIRAP asks the firms it works with to govern their AI use responsibly. We hold ourselves to the same standard.
Where we use AI
| Activity | What AI does | What a person does |
|---|---|---|
| Website pages and articles | Drafts and edits text from NIRAP's notes and sources | Initiates the page, reviews the draft including every fact against its source, and approves publication |
| White paper summaries and guided tours | Drafts a shorter version of a paper NIRAP has already published | Initiates the summary, checks it against the paper, and approves publication |
| Research | Finds and summarizes laws, guidance and studies | Opens and reads the primary source before NIRAP cites it |
| V-AGE Assess drafting assistant | Drafts policies and documents a firm is missing | The firm edits and adopts the documents, and NIRAP reviews the evidence |
| RAICB verification decisions | Nothing. AI does not decide whether a firm is verified | A NIRAP reviewer checks the firm's evidence against the V-AGE standard and decides |
What the label means
The label as it appears on NIRAP pages.
When a page carries the AI-assisted, Human-led label, AI tools helped produce it and people at NIRAP led the work. Human-led means three things happened:
- A person at NIRAP initiated the work.
- A person at NIRAP reviewed the output, including checking its facts against their sources. The person who initiated the work may do the review or assign it to someone else at NIRAP.
- A person at NIRAP approved it for publication.
No one has to have written the first draft. Each labeled page shows the date it was approved, and NIRAP keeps a record of who initiated, reviewed and approved it.
If any of those three steps did not happen, the page does not carry the label.
The label does not guarantee that a page is free of mistakes or errors. It means a person applied due diligence to verify the output. People make mistakes too, which is why we ask readers to report them.
The label is NIRAP's own disclosure. It is not a certification, and NIRAP does not issue it to other organizations.
How review works
A person at NIRAP reviews AI-assisted material before NIRAP publishes it. The reviewer checks four things:
- Every figure, date and legal citation matches its primary source.
- Every quotation is accurate and attributed.
- The page states NIRAP's own views as NIRAP's, and does not attribute them to a regulator or association.
- Nothing on the page claims more than its sources support.
Each labeled page shows the date it was last reviewed. When a source changes, such as a penalty amount that adjusts each year, we review the page again.
Tools and data
These are the AI tools NIRAP currently uses for published material.
| Tool | Used for | Data setting |
|---|---|---|
| Claude, by Anthropic | Drafting, editing and research | A Team or Enterprise plan, whose business terms do not use our conversations to train models |
| ChatGPT, by OpenAI | Drafting, editing and research | A personal plan with model training turned off |
NIRAP staff do not enter a client firm's confidential information into general-purpose AI tools such as Claude or ChatGPT.
Errors and corrections
If you find a mistake on any NIRAP page, email contact@nirap.org with the page address and what is wrong. We review every report.
When we correct a factual error, we add a dated correction note to the bottom of the page. We fix typos and wording without a note.