This article is part of the Iris Ethics Knowledge Base; you can find a directory of all articles here.
Overview
Synthetic data is data generated by an AI system rather than collected directly from human participants. It is designed to preserve the statistical properties of a source dataset while removing directly identifying information about the real individuals whose data informed it. Generating synthetic data does not remove the ethical obligations attaching to the source data used to produce it.
When synthetic data generation involves human research
Where synthetic data is generated using real participant data as a training or reference set, the generation process is itself a use of human research data and falls within the scope of ethical review. The National Statement's requirements for data management apply to the source data regardless of whether the eventual output is synthetic (NS 3.1.43-3.1.46). Researchers should describe, in the data management plan, how the source data will be accessed and processed by the AI tool, and how it will be retained or disposed of once generation is complete (NS 3.1.44).
Where the source data was collected under a consent that did not anticipate its use to generate synthetic data, this is a new use of that data (see Secondary use of data). Depending on the terms of the original consent, this may require a consent amendment, may already be covered if extended or unspecified consent was obtained at collection (NS 2.2.14-2.2.16), or may require a waiver of consent from the review body (NS 2.3.9-2.3.10).
Ethics considerations specific to synthetic data
- Re-identification risk: synthetic data is not inherently non-identifiable. The National Statement treats identifiability as a continuum affected by contextual and technical factors, including the capacity of anyone who receives the data to combine it with other information (NS 3.1, Element 4). Depending on the generation method and the granularity of the source data, synthetic records may correspond closely enough to real individuals to create re-identification risk, particularly for small or distinctive populations. Researchers should assess and document this risk as part of the data management plan (NS 3.1.44).
- Representational validity and bias: synthetic data reproduces the properties, including the biases, of its source data. Research relying on synthetic data to represent a population must address this limitation as part of demonstrating that the methods used are appropriate to the research aims (NS 1.1(b)). Where the source data under-represents particular groups, generated data will do the same, and findings drawn from it should be qualified accordingly.
- Disclosure to participants: where a participant's own data is used to generate synthetic data, this should be disclosed to them as part of the consent process (NS 2.2.2, NS 2.2.6(f)) or, where the use was not anticipated at the time of original consent, through an amendment to the ethics approval.
- Reporting integrity: findings or outputs based on synthetic data should be clearly labelled as such in dissemination materials. Presenting synthetic data as equivalent to directly observed data, without qualification, is inconsistent with the honest and accurate reporting that integrity requires (NS 1.3(c)).
Third-party AI tools and data governance
Synthetic data is often generated using third-party AI platforms. Where client or participant data is provided to such a platform, the data management plan should address data ownership, the location in which the data is processed, and whether the platform retains or uses the data to train its own models (NS 3.1.44). These arrangements should be settled before any data is shared, consistent with the requirement that custodianship, storage, retention, and access arrangements be agreed among collaborating researchers (NS 3.1.43).
What reviewers look for
Reviewers assessing a proposal that generates or uses synthetic data look for a clear description of the source data and its provenance, an assessment of re-identification risk proportionate to the sensitivity of that source data, and a plan for disclosing the use of synthetic data to participants and, where relevant, to readers of any published findings.
This page was last updated on 24 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.