This article is part of the Iris Ethics Knowledge Base; you can find a directory of all articles here.
Overview
AI tools are increasingly used to analyse research data, including natural language processing of interview transcripts, sentiment analysis of open-ended survey responses, and machine learning classification of participant-generated content. These uses are addressed through the National Statement's general provisions on privacy and identifiability (NS 3.1, Element 4), data management (NS 3.1.43-3.1.46), consent (NS 2.2), and research integrity (NS 1.3); the National Statement does not treat AI analysis as a distinct category with its own rules.
Privacy and third-party processing
Where AI analysis is performed by a third-party tool, including a cloud-hosted platform, researchers are disclosing participant data to that provider. This must be addressed in the data management plan, including where the tool stores and processes data, whether it retains data to train its own models, and what security arrangements apply (NS 3.1.44). Because identifiability is a continuum affected by who has access to data and what other information they hold, transferring data to a third-party AI system can increase re-identification risk even where the data was not previously considered identifiable (NS 3.1, Element 4). Data provided to AI systems should be limited to what is necessary for the research, consistent with the general obligation to design research that minimises risk (NS 1.7(a)).
Consent disclosure
Participants must be informed that their data may be analysed using AI tools, including, where applicable, that it may be processed by third-party systems, potentially outside Australia. This forms part of the sufficient information a participant needs to make a voluntary decision to participate, and should address how privacy and confidentiality will be protected (NS 2.2.2, NS 2.2.6(f)).
Algorithmic transparency and research integrity
Researchers should be able to explain, at a general level, how an AI analysis tool works and how it generated its findings. Integrity requires that research be conducted with honesty and that results permit scrutiny (NS 1.3(c),(e)); an analysis process that cannot be explained or audited undermines both. Researchers should maintain an audit trail linking AI-generated findings back to source data, so that a claimed finding, such as a theme identified across transcripts, can be verified against the underlying material rather than accepted on the AI system's authority.
A particular risk is AI hallucination: large language models can fabricate quotes, patterns or themes that do not exist in the source data, presenting them confidently as fact. Human validation by someone familiar with the data is required to detect this and should be documented as part of the analysis process.
Bias and validity
AI analysis tools may reflect biases in their training data that affect the validity of findings, particularly where a tool was trained predominantly on North American or European data and is applied to multicultural, regional, or Aboriginal and Torres Strait Islander contexts. Researchers should assess this risk and address it in the research design, consistent with the requirement that methods be appropriate to the research aims (NS 1.1(b)) and that respect for participants is not compromised by the way the research is conducted (NS 1.1(d)).
What reviewers look for
Reviewers look for a data management plan that addresses third-party AI processing, a clear statement of what will be disclosed to participants, and evidence of human validation of AI-generated findings.
Related resources
This page was last updated on 16 September 2026.
Was this helpful? We'd appreciate your feedback!
This article makes reference to the National Statement on Ethical Conduct in Human Research 2025. It is intended as practical guidance and does not constitute legal advice.
AI Disclosure: Resources in this library were prepared using Large Language Models with input from Iris Ethics staff who guided the scope and design. Subsequent revisions and final versions were developed and approved by Iris Ethics staff.