01 · The Question
Can Ethics Approval Limit How You Use AI?
Your university permits generative AI. Your research team wants to use it to transcribe interviews, analyse participant responses, generate personalised study materials, or introduce an AI tool halfway through an approved project. Can the research ethics committee still say no?
Potentially, yes. Research ethics review is concerned with the rights, welfare, dignity, privacy, and interests of research participants. An AI tool can change how participant information is processed, who or what receives access to it, what risks participants face, and whether the study still operates as described in the approved protocol and consent materials.
The important distinction is that an ethics committee does not necessarily regulate every use of AI by every researcher. Its authority and procedures depend on the jurisdiction, institution, type of research, and applicable ethics framework. The central question is whether the proposed AI use affects research that falls within its oversight.
03 · What You Need to Know
AI Can Change the Ethical Profile of a Research Study
Ethics Committees Review Research, Not Technology in the Abstract
A research ethics committee is not usually deciding whether generative AI is morally acceptable in general. It evaluates the proposed research and whether participants are adequately protected under the ethical and regulatory framework governing the study.
This distinction matters. Using AI to correct the grammar of an administrative email may have little or no bearing on an approved participant protocol. Uploading participant interview transcripts to an external AI provider for analysis is different because it can alter data processing, confidentiality, privacy, and potentially the information on which participants gave consent.
The relevant question is therefore not simply "Does the ethics committee allow ChatGPT?" It is "Does this particular AI use change anything the ethics committee needs to evaluate?"
AI May Introduce New Privacy and Confidentiality Questions
Participant information entered into an external AI system may leave the research environment originally described to the ethics committee. Depending on the service and institutional arrangement, questions may arise about data retention, access, international transfer, secondary use, security, re-identification, provider terms, or whether information is used to improve the system.
The World Health Organization's 2026 report on ethics review and oversight of AI-related health research identifies privacy, data protection, informed consent, transparency, bias, fairness, accountability, and other ethical issues as relevant to oversight of AI-related research. WHO also recognises that AI-related research may challenge existing oversight systems and create risks that conventional review processes need to address.
These concerns do not mean that participant data can never be processed with AI. They mean that the ethical acceptability of doing so depends on the data, system, purpose, safeguards, participant expectations, and applicable requirements.
Watch Out
De-identification is not a magic word. Removing names does not necessarily eliminate re-identification risk, particularly in rich qualitative datasets, rare populations, free-text responses, images, audio, genomic information, or datasets containing unusual combinations of attributes. Evaluate the actual information being processed and the safeguards surrounding it.
Informed Consent May Need to Reflect the AI Workflow
Consent is not necessarily satisfied merely because participants agreed that their information could be "used for research." Whether AI processing needs to be described specifically depends on the nature of the study, applicable requirements, materiality of the processing, foreseeable risks, and what participants were told about data handling.
If researchers tell participants that recordings will remain on secure institutional servers and be accessed only by the research team, later uploading those recordings to an external AI service can be difficult to reconcile with that description.
Likewise, if an AI system directly interacts with participants, generates recommendations affecting them, or materially influences an intervention, the committee may need sufficient information for participants to understand relevant aspects of the research and its risks.
An AI Tool Can Change the Study's Risk Profile
AI can introduce risks beyond confidentiality. Depending on the study, these may include inaccurate outputs, biased classifications, discriminatory effects, inappropriate automated recommendations, opacity in decision-making, cybersecurity risks, or unequal performance across participant groups.
WHO's current guidance on AI-related health research draws attention to fairness and inequity as well as risks that may be particularly consequential in low- and middle-income countries, including data colonialism, power imbalances, ethics dumping, and inequitable benefit sharing.
An ethics committee may therefore ask not merely whether the AI system "works," but how its limitations could affect participants and what safeguards are in place when it fails.
The Committee May Ask Who Remains Accountable
Delegating a task to an AI system does not eliminate researcher responsibility. If AI is used to screen participants, classify sensitive responses, generate an intervention, detect risk, or support decisions affecting participants, the protocol may need to explain who checks the output and what happens when the system produces an error.
A statement such as "the AI will decide" is unlikely to answer the underlying ethics question. The committee may reasonably need to understand the human oversight, escalation procedures, error handling, validation, and safeguards built into the study.
AI Use Can Be Part of the Methodology
Sometimes AI is merely a research aid. In other studies it becomes part of the method itself.
| AI Use |
Possible Ethics Relevance |
| Grammar correction of non-sensitive researcher notes |
May have little direct participant-protection relevance, subject to institutional rules |
| Transcribing participant interviews |
Confidentiality, data transfer, accuracy, consent, security |
| Coding qualitative participant data |
Data processing, confidentiality, methodological validity, bias, human verification |
| AI chatbot interacting with participants |
Disclosure, participant understanding, safety, inappropriate responses, data collection |
| AI determining participant eligibility |
Fairness, error, exclusion, transparency, oversight |
| AI-generated personalised intervention |
Risk, reliability, human oversight, potential harm, consent |
| AI analysis of health or genomic information |
Privacy, sensitivity, re-identification, security, bias, secondary use |
The more AI affects participant-facing procedures, sensitive information, eligibility, interventions, or consequential decisions, the more likely it is to raise substantive ethics-review questions.
Introducing AI After Approval Can Be a Protocol Change
This is one of the most important practical points for researchers. Ethics approval generally applies to the research as reviewed, not to any method researchers later decide to substitute.
Under the U.S. HHS human-subject protection framework, for example, institutional review boards must have procedures ensuring that proposed changes to approved research are reported and that changes are not initiated without IRB review and approval except where necessary to eliminate apparent immediate hazards to participants. HHS guidance similarly tells investigators seeking to modify ongoing IRB-approved research to obtain approval before implementing the proposed modification, subject to that emergency exception.
The exact procedure differs across jurisdictions and institutions, so this should not be treated as a universal U.S.-style amendment rule. The broader lesson is portable: if your approved protocol described one data-processing or participant procedure and you want to introduce AI in a way that materially changes it, check with the responsible ethics body before making the change.
AI use already described and approved
The committee has evaluated the proposed AI activity as part of the submitted research, subject to the conditions of its approval.
AI use introduced later
The researcher may need to determine whether the new activity constitutes a modification requiring review before implementation.
Ethics Approval Does Not Replace Other AI Requirements
A research ethics committee may approve an AI-assisted procedure from the standpoint of participant protection while another institutional authority prohibits the chosen service for security reasons. Conversely, an AI platform approved by the university's IT department is not automatically approved for use with participants in a particular study.
This is why researchers need to understand the broader set of policies that can govern AI use in research. Ethics review is one layer, not a universal compliance certificate.
The Ethics Committee May Approve, Require Changes, or Decline the Proposed Use
Where an ethics body has authority over the research, review is not limited to a binary choice between approving everything and banning AI entirely.
For research subject to the U.S. HHS framework, for example, IRBs can approve research, require modifications to secure approval, approve under appropriate conditions, or disapprove research that does not satisfy applicable approval criteria. The same general idea appears in many ethics-review systems even though terminology and legal authority vary.
An AI-related condition might concern data minimisation, participant information, consent language, human oversight, exclusion of certain data from the AI system, use of an institutionally approved platform, or a revised procedure for handling problematic outputs.
07 · A Quick Checklist
Before Using AI in Human-Participant Research
Before implementing the AI use, check:
Identify exactly what role AI will play in recruitment, interaction, data collection, processing, analysis, intervention, or decision-making.
Map what participant information will enter the AI system and whether any of it remains identifiable or sensitive.
Check how the system handles submitted information, including relevant access, retention, security, and secondary-use arrangements.
Compare the AI workflow with what participants are told in the consent process and participant information materials.
Assess potential errors, bias, unfair outcomes, inappropriate outputs, and other participant risks relevant to the proposed use.
Specify meaningful human oversight when AI outputs could affect participants or consequential research decisions.
For an approved study, determine whether adding or changing AI use requires ethics review or a protocol amendment before implementation.
Check institutional privacy, security, data-management, legal, contractual, and funder requirements in addition to ethics approval.
