03 · What You Need to Know
How to Determine Whether AI Use Requires Research Ethics Review
Begin With the Scope of the Research Ethics Framework
Research ethics review is governed by the requirements applicable to the study, institution, and jurisdiction.
Human-participant research commonly requires assessment of matters such as informed consent, participant welfare, confidentiality, equitable selection, and the relationship between anticipated risks and potential benefits.
However, not every scholarly activity involving information about people is necessarily subject to the same review procedures.
Some activities may be exempt, excluded from a particular regulatory definition, or eligible for different levels of review. The terminology and criteria vary across regulatory systems.
Researchers should therefore begin by determining whether the proposed study falls within the relevant ethics review framework, rather than assuming that the presence of AI alone decides the matter.
Does AI Automatically Trigger a New Ethics Review?
No universal rule requires a separate ethics application every time a researcher uses generative AI.
For example, using AI to improve the readability of a manuscript without submitting restricted information or changing scientific content may not affect an existing human-participant protocol.
By contrast, introducing an AI chatbot that communicates directly with participants may change the procedures, interactions, and foreseeable risks described in the protocol.
The decisive consideration is what AI actually does within the study.
Researchers should distinguish peripheral assistance from AI activities that alter participant experiences, data processing, research measurements, or consequential decisions.
Which AI Activities Are Most Likely to Warrant Ethics Review?
| AI-assisted activity |
Potential ethical concern |
Review consideration |
| AI-generated manuscript wording |
Scientific accuracy and publication integrity |
Usually governed by author and publication requirements unless additional research risks arise |
| AI-assisted recruitment |
Participant selection, targeting, privacy, and fairness |
Determine whether recruitment procedures or information processing change |
| AI-supported consent interactions |
Accuracy of explanations and participant understanding |
Assess whether the approved consent procedure is modified |
| AI-delivered intervention |
Unpredictable responses and potential participant harm |
Evaluate the intervention and associated safeguards through applicable review |
| Automated eligibility screening |
Incorrect inclusion, exclusion, or safety decisions |
Assess validation, oversight, and protocol requirements |
| AI processing of confidential interviews |
Confidentiality, disclosure, retention, and reidentification |
Determine whether the processing is authorized and requires an amendment |
| AI-generated research measurements |
Measurement validity, classification error, and potential bias |
Assess methodological and ethical implications of the procedure |
| AI analysis of public, non-identifiable materials |
Scientific validity and any remaining ethical obligations |
Determine whether the activity falls within the applicable review framework |
These are illustrative review considerations, not a universal list of activities that automatically require formal ethics approval.
When Does AI-Assisted Recruitment Require Review?
AI can assist with drafting recruitment messages, identifying potential participants, or prioritizing invitations.
Drafting a message for human review is different from allowing a model to access student records and determine who should be contacted.
Automated recruitment may affect equitable participant selection, privacy, and compliance with approved eligibility criteria.
For example, an AI system might disproportionately prioritize individuals whose digital records are more complete, thereby introducing selection bias.
Researchers should examine whether the proposed process changes the approved recruitment strategy and whether the underlying information may lawfully and ethically be processed.
Where required, the revised recruitment procedure should be reviewed before implementation.
What if Participants Interact Directly With AI?
Participant-facing AI introduces concerns beyond ordinary research administration.
An AI chatbot may generate variable responses, misunderstand questions, provide inappropriate information, or behave differently across participants.
These characteristics may affect the research intervention, participant experience, and the comparability of experimental conditions.
Researchers should define the intended interaction, foreseeable failure modes, safeguards, and procedures for responding to harmful or unexpected outputs.
Where AI is part of the experimental intervention, its role should be described in the research protocol and evaluated under the applicable ethics framework.
The ability of a model to produce fluent conversation does not establish that the interaction is safe or ethically appropriate.
Can AI Be Used in the Informed Consent Process?
AI may help researchers prepare plain-language explanations or translate draft consent materials for qualified review.
However, an AI system that independently answers participant questions during consent may introduce additional risks.
The system could provide incomplete explanations, misunderstand a question, or introduce statements inconsistent with the approved materials.
Researchers should determine whether the AI system forms part of the consent procedure and how participants can obtain accurate answers from responsible personnel.
Consent is not established merely because information was delivered. The process must satisfy the applicable requirements for meaningful, voluntary, and adequately informed participation.
When Does AI-Assisted Eligibility Screening Raise Ethical Concerns?
Eligibility screening may determine whether individuals can participate in a study or receive a particular intervention.
An incorrect AI classification could exclude eligible participants or include individuals for whom participation presents additional risks.
Researchers should establish the approved criteria, relevant validation evidence, and procedures for reviewing uncertain or consequential recommendations.
For sensitive interventions, qualified human oversight may be necessary under the applicable protocol.
AI should not independently redefine eligibility criteria or override established safety requirements.
Where automated screening materially changes participant selection or risk, the appropriate ethics review should occur before deployment.
Does AI Transcription Require Ethics Approval?
AI-assisted transcription does not automatically require a separate ethics application in every research setting.
The relevant concern is whether the transcription arrangement changes the approved processing of participant recordings.
Voice recordings may contain directly identifiable information, sensitive statements, and contextual details that permit identification.
Using an external transcription provider may introduce additional access, storage, retention, or international processing arrangements.
Researchers should establish whether these arrangements are permitted under the approved protocol, consent conditions, and applicable institutional requirements.
Where the proposed processing represents a material change, the responsible ethics authority should determine whether an amendment is necessary.
What if the Data Have Been De-Identified?
De-identification may reduce privacy risks, but its effectiveness depends on the information and circumstances.
Removing names does not necessarily eliminate identifying details in interview narratives or linked datasets.
Researchers should distinguish de-identification from anonymization and assess the possibility of reidentification.
Even when information is no longer identifiable under an applicable standard, contractual, institutional, or consent-related restrictions may remain relevant.
The absence of direct identifiers does not automatically establish that any AI processing arrangement is permissible.
When Does AI-Assisted Analysis Require an Ethics Amendment?
Introducing AI during analysis may or may not require a protocol amendment.
If the system is used only to assist with code development using synthetic examples, the change may have limited implications for participant protection.
If confidential transcripts are transferred to an external service or an AI-generated classification changes participant-related procedures, the implications may be substantially different.
Researchers should examine whether the proposed activity changes the approved methodology, information handling, participant risks, or consent arrangements.
When the answer is uncertain, the responsible ethics office should provide the applicable determination.
What Do U.S. Regulations Say About Changes to Approved Research?
For research subject to the relevant U.S. Department of Health and Human Services Common Rule requirements, 45 CFR 46.108(a)(3)(iii) addresses institutional review board procedures for reviewing changes to approved research before implementation.
The provision recognizes an exception for changes necessary to eliminate apparent immediate hazards to participants.
This is a jurisdiction-specific regulatory requirement, not a universal rule governing every research institution worldwide.
Researchers should consult the applicable regulations and institutional procedures before introducing material changes.
They should also distinguish an IRB's determination of review requirements from the researcher's own informal assessment.
What Should Researchers Consider in the Philippines?
In the Philippines, researchers should consider the requirements of their institutional research ethics committee, relevant national research ethics guidance, and applicable data protection obligations.
The Philippine Health Research Ethics Board provides national guidance relevant to health-related research ethics review.
Republic Act No. 10173, or the Data Privacy Act of 2012, establishes obligations concerning personal information processing. However, the law should not be interpreted as imposing a universal AI-specific research ethics review requirement.
For educational research involving identifiable student records, institutional research ethics procedures and data protection requirements may both be relevant.
Researchers should seek case-specific guidance rather than assume that requirements applicable to one type of study automatically apply to another.
How Does Ethics Review Differ From Privacy and Security Review?
Research ethics review
Evaluates research activities and participant protections under the applicable ethical and institutional framework.
Privacy and information-security review
Evaluates whether information processing, access, storage, transfer, retention, and safeguards satisfy relevant obligations.
The two forms of review can overlap, but neither automatically replaces the other.
An AI service may satisfy institutional security requirements while remaining unsuitable for an unapproved participant-facing intervention.
Conversely, an approved study may still require additional assessment before transferring confidential data to a new provider.
Does Ethics Approval Establish That an AI Method Is Scientifically Valid?
No. Ethical authorization and methodological validation are distinct responsibilities.
An ethics committee may evaluate whether a proposed research procedure provides appropriate participant protections.
That determination does not necessarily establish the accuracy of an AI classifier, the correctness of generated code, or the validity of a research measurement.
Researchers should independently evaluate whether the proposed AI system is suitable for its scientific task.
For example, an authorized sentiment classifier may still perform inadequately on multilingual student responses.
Ethics approval should not be treated as evidence that every analytical output is reliable.
What Should an AI-Related Ethics Submission Explain?
The required documentation depends on the institution and research design.
Relevant information may include the intended AI task, the system or provider, the information processed, participant interactions, foreseeable risks, and proposed safeguards.
For consequential AI-assisted decisions, researchers may also need to explain validation, human oversight, error handling, and procedures for responding to unexpected outputs.
Where information is transferred externally, the submission may need to identify relevant processing and confidentiality arrangements.
The objective is to explain the actual procedure rather than merely state that AI will assist the research.
When Should Researchers Seek an Ethics Determination Rather Than Assume Exemption?
Researchers should seek an institutional determination when the applicable rules are unclear, particularly for new participant-facing activities, changes to approved procedures, or processing arrangements involving sensitive information.
An activity may qualify for exemption or another form of review, but researchers should follow the institution's established process for making that determination.
Self-declaring an activity exempt may be inconsistent with institutional requirements.
Where an AI use is introduced after approval, the research team should establish whether the change can proceed under the existing protocol.
Advance clarification is generally preferable to attempting to justify an unreviewed material change after it has occurred.
When Should Researchers Avoid the Proposed AI Use?
Researchers should not proceed with an AI activity when necessary authorization is unavailable, when participant risks cannot be adequately controlled, or when the processing would violate applicable obligations.
The team may need to modify the procedure, select an authorized environment, obtain additional validation, or use an alternative method.
For example, confidential interviews should not be transferred to an external service merely because it offers faster analysis if the proposed processing is not authorized.
Ethical research requires that the anticipated benefits of AI assistance be considered alongside the rights and welfare of the people whose information or participation makes the research possible.
