Subfractal
Set the rules

What NATA Has Said About AI in Accredited Facilities, and What It Hasn't

Everything we could find that NATA has published on AI, checked on 22 October 2026, including its new six questions, which are commentary rather than criteria. Then what the UK tells its labs, and the notification question that is still open.

For: Quality managers, laboratory owners and authorised representatives at NATA-accredited facilities

If you are responsible for a NATA accreditation, you have probably been asked some version of this in the last year: are we allowed to use AI?

We could find no NATA document that gives a one-line answer. Below is everything we could find that NATA has said, in date order, checked against NATA's own pages on 22 October 2026. Then the UK, then the gaps. If we have missed something, tell us and we will correct it.

What has NATA said about AI so far?

When Where What
February 2024 Member Advisory Forum minutes The minutes record that "it needs to be managed like any other tool", that it should "complement human efforts rather than replace them", and that confidentiality and software verification or validation should be prioritised.
November 2024 NATA news NATA's Chief Information Officer said that, as AI is adopted, assessments by accreditation bodies "will need to have an increased focus on requirements for verification and validation, control of data, and organisational governance and accountability".
February 2025 NATA news Mostly about NATA's own adoption. It warns over-reliance could reduce the role of human judgement, and says final decisions "should still involve human oversight".
March 2025 NATA news NATA announced it had joined the international AI Quality Infrastructure (AIQI) Consortium.
November 2025 Member Advisory Forum minutes Updates on NATA's internal AI policy and on "a working group established to examine how to assess AI used by conformity assessment bodies".
September 2026 NATA news Responsible AI use by NATA's own staff, guided by ISO/IEC 42001. Its plainest line: "accountability always remains with the person using the technology".
September 2026, effective October 2026 ISO/IEC 17020 application document, clause 6.2.9 The first accreditation criteria we found that name AI: "Facilities are responsible for any technology used in inspections including computers, automated equipment, data processing, AI, augmented reality, or remote inspection techniques."
October 2026 Annual Report 2026 NATA represents Australia on AI standards committees, including work on ISO/IEC 42007, an upcoming standard for accredited testing, inspection and calibration of AI systems.
13 October 2026 NATA news "Can you Trust an AI-Supported Result?" Commentary, not criteria: six things organisations "should consider" before relying on an AI-supported result, which "should not be treated as a formal NATA checklist, but as useful questions to support responsible decision-making".

Read together: AI is a tool. Validate it. Control the data. A person stays accountable.

Only one row is a requirement. Clause 6.2.9 of the 17020 application document applies to inspection bodies that use technology in their inspections, AI included: the technology must be suitable and adequate, data integrity and security procedures must be in place, and it must be revisited "whenever changes could affect inspection outcomes". We cover it in Your inspection body now needs a written position on AI.

What did NATA's October 2026 article add?

It is the closest NATA has come to speaking to accredited facilities directly, but it is a news article, not accreditation criteria. Its six questions cover purpose, data, validation, limitations, oversight and accountability. Is "a competent person" reviewing its outputs? Is it "clear who remains responsible for the final result"? It also notes that ISO/IEC 42001 addresses governance and does not certify that an individual AI application will be accurate.

Two points matter. First, NATA's article frames the question as one of influence, not use:

For accredited facilities, the critical issue is not simply whether AI is being used. It is whether that use could influence an accredited activity or outcome.

Second, it says the principles behind the questions "are already in existing accreditation standards like ISO/IEC 17025", without saying which clauses. It ends: "AI can support a result. It cannot take responsibility for it."

What hasn't NATA said?

No AI in the criteria for testing and calibration laboratories. We could find no mention of AI in NATA's application document for ISO/IEC 17025, dated October 2024. It does say electronic reports must "have been appropriately authorised for release", with controls over how results and reports are generated, accessed, stored and backed up.

No guidance on how to adopt it. We could find no NATA guidance on how to validate an AI tool, what records to keep, or which uses it regards as high risk. The October article asks the questions; it does not say what a good answer looks like.

Nothing yet on what assessors will ask. NATA's 2026 annual report credits more than 2,500 volunteer technical assessors. The November 2025 minutes record a working group on how to assess AI used by accredited bodies. We could find no published output from it.

No answer on notification. This is the practical one, and it needs its own section.

What has the UK's accreditation body said?

UKAS, the UK's accreditation body, published a technical bulletin on AI in accredited conformity assessment in June 2025, and a second, jointly with its German counterpart DAkkS, in March 2026. Three points matter here.

They say it is not a new requirement. The June 2025 bulletin says its intention is not "to establish new or additional requirements for accreditation, but to contextualise existing requirements".

They expect validation. Bodies should validate AI systems "according to a documented procedure", and work out the impact when a validated system changes. If the change affects the original validation, it is validated again. The joint bulletin adds that final decisions on conformity must not be delegated to an AI system under current accreditation frameworks.

They expect to be told. UKAS reminds bodies to tell it of significant changes, then says: "The adoption or implementation of AI technologies is considered a noteworthy development." Bodies are "therefore required to notify UKAS". The joint bulletin's page lists what a notification should cover, as a minimum:

  1. a brief description of the system, its purpose, the training data or knowledge used, and the implementation approach;
  2. where it sits in the conformity assessment process;
  3. the degree of reliance on it, from administrative support to decision support to any level of automated decision-making.

None of that binds a NATA-accredited facility. It is the clearest published view we found of how an accreditation body thinks about it.

Do you have to notify NATA within 14 days if you start using AI?

NATA already has a notification rule. Its criteria on the responsibilities of authorised representatives say NATA must be notified in writing within fourteen days of "a significant change in the persons, ownership, other resources, management practices, premises or functions of the accredited facility", or in the personnel authorised to release results. (NATA Rules R.27 still calls those people "approved signatories", the older term many people search for.) R.27 frames the trigger as a change "which could impact on the member's ability to satisfy the conditions for the continuance of accreditation".

Is an AI tool a significant change in resources or management practices? We could find no NATA guidance that answers it. UKAS has answered the same question for its own bodies with a yes. A spell-checker plainly is not significant. An example, made up: a calibration laboratory starts using a tool that drafts the commentary on its certificates. That is much harder to wave through, and it is exactly the kind of use NATA's October article says could influence an accredited outcome.

One way to stop guessing is to ask. That is your call, not ours. A short email to your NATA contact, describing what the tool does and where it sits, puts the answer on file. If the reply is "no need to tell us", you have that in writing.

What can a careful facility do this month?

  1. Find out what is already in use. Ask staff, without blame. Include AI features inside software you already own.
  2. Sort each use into three piles. Administration that touches no client data or results. Support for a person's decision. Anything that produces or changes a result or a report. The piles are ours, loosely following the UKAS and DAkkS scale of reliance, not NATA's categories.
  3. Treat the second and third piles like any other software that affects results. A written purpose, a check against known cases before use, a record of the version, and a re-check when it changes. NATA's six questions are a useful test of coverage.
  4. Check where the data goes. The February 2024 forum minutes named confidentiality as a priority. Client results typed into a public tool have left the building.
  5. Make sure the record shows the person. Whoever authorises a result should be able to say why, and the record should show that they did. For what an assessor is likely to probe, see If an AI drafts it and you sign it.
  6. Consider asking NATA the notification question in writing, and keep the answer with your records.

One question

We are asking people in accredited facilities how they are handling this. No pitch: has anyone in your facility started using AI for anything, and did you have to decide whether to tell NATA? We will share what we hear, without names.

Write to warwick@subfrac.com.

Sources

  1. Member Advisory Forum minutesNATA, February 2024
  2. NATA assessing opportunities with AINATA, 6 November 2024
  3. Part 2 – NATA: Assessing the adoption of AINATA, 3 February 2025
  4. Announcing NATA membership of the AI Quality Infrastructure (AIQI) ConsortiumNATA, 4 March 2025
  5. Member Advisory Forum, 9th meeting minutesNATA, 11 November 2025
  6. Responsible AI at NATANATA, 8 September 2026
  7. General Accreditation Criteria: ISO/IEC 17020 Standard Application DocumentNATA, Issued September 2026, effective October 2026
  8. Accreditation Matters: NATA Annual Report 2026NATA, October 2026
  9. Can you Trust an AI-Supported Result?NATA, 13 October 2026
  10. General Accreditation Criteria: ISO/IEC 17025 Standard Application DocumentNATA, October 2024
  11. General Accreditation Criteria: Responsibilities of Authorised RepresentativesNATA, September 2024
  12. NATA RulesNATA, September 2026
  13. The use of Artificial Intelligence (AI) technologies in accredited conformity assessmentUKAS, 20 June 2025
  14. Understanding AI in accredited conformity assessment: AI Technical Bulletins (joint UKAS–DAkkS bulletin)UKAS, 6 March 2026

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

Warwick (Woz) Heathwood is a co-founder of Subfractal.