ISO/IEC 17020:2026, Clause 6.2.9: Why Your Inspection Body Should Write Down Its Position on AI
Since 1 October 2026, NATA has assessed inspection bodies against the new edition of ISO/IEC 17020, and its criteria now name AI. Here is what the clause asks of a body that uses AI, and why writing down your position is the sensible way to show you meet it.
For: Technical and quality managers at NATA-accredited inspection bodies
If you run an accredited inspection body in Australia, the standard you are assessed against changed this year. One of the changes uses a word that NATA's inspection criteria have not used before: AI.
This is a plain account of that change, taken from NATA's published documents. We have not read the standard itself; where we describe it, we are going on NATA's summary of it. Buy the standard, read NATA's current criteria for yourself, and ask your NATA client coordinator where anything is unclear.
What changed, and when?
ISO/IEC 17020:2026 was published in March 2026. It replaces the 2012 edition. NATA's transition policy sets out the timetable:
- From 1 October 2026, all NATA assessments of inspection bodies are conducted against the 2026 edition.
- Your accreditation converts at an assessment. NATA says the changes are large enough that an assessment is needed. That happens at your next scheduled visit after 1 October 2026, or earlier if you ask, as a chargeable variation.
- Before that visit, you complete NATA's ISO/IEC 17020:2026 Implementation Checklist and supply information showing how you comply, as part of the usual arrangements before an on-site visit.
- Shortfalls become findings, written up and responded to in the usual way before accreditation continues under the new edition.
- 31 March 2029 is the end of the transition. NATA's policy says facilities that have not transitioned by then will have their accreditation suspended.
Two other NATA documents are free to download and worth an afternoon. The gap analysis sets the two editions side by side. The Implementation Checklist lists the clauses with new or major changes. For each one, you record the action you took and attach supporting evidence. An authorised representative signs a declaration at the front. Clause 6.2.9 is on that list.
What does clause 6.2.9 actually say?
In the 2012 edition, the equivalent clause was 6.2.13, and it dealt with computers and automated equipment. NATA's gap analysis rates the change as major. It describes the new clause as widening the idea of technology to take in data processing, artificial intelligence, augmented reality and remote inspection techniques.
NATA's gap analysis summarises the new clause like this:
If the inspection body uses technology, in connection with inspections (e.g. computers or automated equipment, data processing, artificial intelligence, augmented reality or remote inspection techniques), it shall ensure that: a) the technology is suitable and adequate for use; b) procedures established and implemented for protecting the integrity and security of data; c) equipment, including hardware and software, is maintained in order to ensure proper functioning.
Note the first word: if. The clause applies to technology you use.
NATA's application document, issued September 2026 and effective October 2026, puts its reading in a few sentences:
Facilities are responsible for any technology used in inspections including computers, automated equipment, data processing, AI, augmented reality, or remote inspection techniques. They must ensure such technology is suitable and adequate, that data integrity and security procedures are in place, and that hardware and software are properly maintained. This applies regardless of how novel the technology is and must be revisited whenever changes could affect inspection outcomes.
What else in the new criteria sits beside it?
Three other parts of the criteria connect to 6.2.9.
Your methods have to cover it. According to NATA's gap analysis, clause 7.2.5 lists what inspection methods and procedures must address. Two items on that list are new in kind: activities where the inspector's judgement may rely on the use of technology, and inspection results from AI or digital developments. A related new clause, 7.2.6, asks for non-standard methods to be validated, including when technology is used. Both 7.2.5 and 7.2.6 are on the Implementation Checklist.
Changes are controlled. NATA's application document says information management systems used in inspection must be validated as suitable for their purpose. It says that software changes, configuration amendments and system updates must go through change control, "including authorisation, documentation and validation, before implementation". It adds that commercial off-the-shelf software may be considered pre-validated when used within its specified application range.
A person still authorises. The same document says all inspection results, including reports issued electronically, must be authorised before the report or certificate is released. We look at what that means for AI-drafted reports in If AI drafts it and you sign it.
Why write it down if you don't use AI?
Strictly, a body that uses no AI in its inspection work has nothing to show under the AI part of 6.2.9. We think it is still worth writing that down, for three reasons.
AI arrives inside software you already have. Report-writing tools, image and video review software, office suites and phone apps have all added AI features, often switched on by an update. A body can start using AI without deciding to.
People use things unofficially. An inspector who pastes field notes into a chat tool on a personal phone to tidy up a report has used AI in connection with an inspection. They have also sent client information somewhere you did not choose.
The criteria expect you to revisit. NATA's application document says the responsibility must be revisited whenever changes could affect inspection outcomes. A position you have never written down is hard to revisit, and hard to show anyone.
So "we do not use AI in inspection work" is a perfectly good position. It needs to be true, checked and on paper. If you do use AI, the written position is how you show you are responsible for it.
What hasn't NATA said?
We could find no NATA criteria or formal guidance on how an accredited facility should adopt or control AI, or on what an assessor will look for under 6.2.9. We searched NATA's site on 22 October 2026.
The nearest thing is a news article NATA published on 13 October 2026, "Can you trust an AI-supported result?". It lists six things an organisation should consider before relying on an AI-supported result: purpose, data, validation, limitations, oversight, and accountability for the final result. NATA says they "should not be treated as a formal NATA checklist, but as useful questions to support responsible decision-making". They are still a good outline for the one-page position in the steps below.
Assessments against the new edition began on 1 October 2026, so there is little shared experience yet. Technical assessors are practitioners from industry, and each will bring their own view. That is a reason to be plain and specific in what you write, and to keep your evidence simple.
Our companion piece, What NATA has said about AI, and what it hasn't, sets out everything we found.
What can you do before your next visit?
None of this needs new software. These are things a careful inspection body might consider; they are not accreditation advice.
- Find your date. Check when your next scheduled visit is. Download the gap analysis and the Implementation Checklist from NATA's website.
- List every piece of technology that touches an inspection. Measuring equipment with firmware, cameras and drones, tablets and apps, report templates, spreadsheets with formulas, and anything that stores or sends inspection data. Then ask two awkward questions. Which of these has an AI feature? And what are people using that is not on the list?
- Write one line per item under five headings. What it is used for. Who is responsible for it. How you know it is suitable. Where the data goes. What happens when it updates. A simple table is enough.
- Write your position on AI on one page. Say what is permitted and what is not. Say who can approve a new use, and how. Say what an inspector does when a tool they already use gains a new feature. Say how often you will review the page, and that any change that could affect inspection outcomes triggers a review.
- Mark where judgement leans on a tool. Go through your inspection methods and mark every step where the inspector's judgement relies on a tool's output: a wall-thickness reading, an automated defect call, a calculated load. Check that the method says what the inspector does with that output, and what they do when it looks wrong.
- Put software updates under change control. For most small bodies this can be a short log: what changed, who authorised it, and how you checked it still does what you rely on it for.
- Draft your checklist response for 6.2.9 and 7.2.5. Point each one at the evidence from steps 2 to 6. If you can write that response in plain sentences, you are probably ready to talk about it on the day.
One question
We are trying to learn how inspection bodies are reading this clause. So, a question with no pitch attached: have you completed the Implementation Checklist yet, and what did you write against 6.2.9?
Write to warwick@subfrac.com. We will share what we hear, without names.
Sources
- Specific Accreditation Criteria: Transition Policy for the implementation of ISO/IEC 17020:2026
- General Accreditation Criteria: ISO/IEC 17020 Standard Application Document
- Specific Accreditation Guidance: ISO/IEC 17020:2026 Gap Analysis
- ISO/IEC 17020:2026 Implementation Checklist
- Can you trust an AI-supported result?
Facts last checked:
Subfractal is not affiliated with or endorsed by NATA, and nothing here has been reviewed by it. This is general information, not accreditation or legal advice.
Who wrote this
Dom O'Carroll is a co-founder of Subfractal.
