03 · What You Need to Know
Which AI Activities Require Advance Research Authorization?
Why Does Approval Matter Before Data Collection?
Research ethics review is intended to evaluate relevant aspects of a study before participants are exposed to its procedures and foreseeable risks.
When AI is introduced into participant-facing activities or data processing, it may change the information collected, the people or organizations accessing it, or the way decisions affecting participants are made.
These changes can have consequences that are difficult to reverse after collection.
For example, once identifiable interview recordings have been transmitted to an unauthorized external service, obtaining subsequent approval does not undo the disclosure.
Advance planning allows researchers to establish appropriate safeguards, explain relevant processing arrangements to participants, and determine whether the proposed use is consistent with the research design.
Ethics Approval, Privacy Review, and Tool Authorization Are Different
Researchers should distinguish the different forms of authorization that may apply.
Research ethics approval
Evaluates research procedures and participant protections under the applicable ethics review system.
Institutional or information-security authorization
Determines whether a particular system and processing arrangement are permitted under organizational requirements.
Privacy or data protection review may additionally examine the lawful and appropriate handling of personal information, including access, retention, disclosure, and security safeguards.
These reviews may overlap, but they answer different questions.
A university-approved AI platform is not automatically approved for every research protocol. Similarly, research ethics approval does not necessarily establish that a particular commercial service meets institutional security requirements.
Researchers should identify which approvals are relevant rather than assuming that one authorization satisfies all obligations.
Which AI Uses Are Most Likely to Require Advance Review?
| Proposed AI activity |
Potential concern |
What should be established before use |
| AI-assisted participant recruitment |
Selection bias, inappropriate targeting, use of personal information |
Approved recruitment procedure and authorized data processing |
| AI-generated consent interactions |
Inaccurate explanations or incomplete understanding |
Approved consent process and appropriate human safeguards |
| Automated eligibility screening |
Incorrect inclusion or exclusion decisions |
Validated criteria, oversight, and applicable protocol approval |
| AI transcription of interviews |
Disclosure of recordings or identifiable speech |
Permitted processing environment and confidentiality safeguards |
| AI analysis of identifiable records |
Privacy, access, retention, and secondary processing |
Appropriate ethics and data-processing authorization |
| AI-delivered research interventions |
Unpredictable responses and participant harm |
Risk assessment, approved procedures, validation, and oversight |
| AI-generated research measurements |
Validity, reliability, and possible classification bias |
Methodological validation and any required protocol approval |
| AI editing of non-sensitive manuscript text |
Altered meaning or inaccurate statements |
Author review and applicable institutional or publication policy |
This table identifies activities that warrant advance assessment. It does not establish that every activity requires the same formal approval or that a particular ethics committee must review all of them.
Does AI-Assisted Recruitment Require Approval?
AI may help researchers prepare recruitment messages, identify potential participants, or prioritize invitations.
Drafting a message for subsequent human review is different from allowing an AI system to access student records and decide whom to contact.
Automated targeting may affect participant selection, fairness, privacy, and the implementation of approved eligibility criteria.
Researchers should establish whether the proposed procedure is consistent with the protocol and whether the information used for recruitment may be processed by the selected system.
Where AI materially changes recruitment procedures, the relevant ethics authority should determine whether review or amendment is required.
Can AI Be Used to Explain Informed Consent?
An AI system may assist with preparing plain-language explanations of research participation.
However, allowing participants to interact directly with a generative system introduces different concerns.
The system may provide incomplete or inconsistent explanations, misunderstand participant questions, or introduce information not contained in the approved consent materials.
Researchers should determine whether the AI interaction forms part of the approved consent process and how participants can obtain reliable answers from qualified personnel.
Where required, the consent procedure and supporting materials should be reviewed before implementation.
AI assistance should not be assumed to establish informed consent merely because participants receive a fluent explanation.
When Does AI Transcription Create a New Approval Issue?
Interview recordings can contain identifiable voices, names, locations, and contextual information.
Using an external transcription service may introduce additional processing and access arrangements beyond those described in the original protocol.
Researchers should determine whether recordings are uploaded, retained, used for model improvement, accessed by subcontractors, or processed in other jurisdictions.
The relevant terms and safeguards should be examined before the recordings are transferred.
An automated transcription service is not automatically inappropriate, but its use must be consistent with the study's confidentiality and authorization requirements.
Transcription accuracy should also be evaluated, particularly where mistakes could alter the meaning of participant statements.
Does Removing Participant Names Eliminate the Need for Review?
No. Removing direct identifiers does not necessarily make research information anonymous.
Interview excerpts may contain distinctive experiences, institutional roles, locations, or other details that permit identification when combined with additional information.
Researchers should distinguish de-identification, pseudonymization, and anonymization according to the applicable legal and methodological context.
The adequacy of the transformation depends on the information, available auxiliary data, and intended processing environment.
Researchers should not assume that replacing names with participant codes automatically authorizes external AI processing.
What if AI Is Used Only After Data Collection?
Introducing AI after collection does not automatically remove ethics or data protection obligations.
The proposed use may still change how participant information is processed or how the approved analysis is conducted.
For example, uploading confidential interview transcripts to a commercial language model may introduce a processing arrangement that was not originally evaluated.
Researchers should consult the applicable institutional procedures before implementing a material change.
However, an AI-assisted analytical activity involving appropriately authorized non-identifiable data may not require a new ethics review in every institution.
The decision should follow the actual requirements governing the study rather than a universal assumption.
What Do U.S. Human-Subjects Regulations Illustrate?
For research subject to the relevant U.S. Department of Health and Human Services regulations, 45 CFR 46.108(a)(3)(iii) requires institutional review boards to establish procedures ensuring that proposed changes to approved research are reviewed and approved before implementation, except where changes are necessary to eliminate apparent immediate hazards to participants.
This provides an example of a jurisdiction-specific requirement rather than a universal rule for all research worldwide.
Researchers should consult the regulations and institutional procedures applicable to their own study.
In particular, the distinction between an administrative adjustment and a substantive change to participant procedures should be determined under the relevant review system.
What Should Researchers Consider in the Philippine Context?
Researchers working in the Philippines should consider the Data Privacy Act of 2012, applicable National Privacy Commission issuances, institutional data protection policies, and the requirements of the relevant research ethics committee.
The Data Privacy Act provides a legal framework for personal information processing, but it should not be interpreted as imposing a universal AI-specific ethics approval requirement.
The appropriate lawful basis, safeguards, and institutional responsibilities depend on the nature of the information and processing activity.
Researchers conducting health-related studies may also be subject to applicable Philippine health research ethics requirements.
When the processing arrangement is unclear, the institution's data protection officer and research ethics office are appropriate sources of case-specific guidance.
Should AI-Generated Measurements Be Approved Before Collection?
If AI will generate primary research variables, its role should ideally be established during methodological planning.
For example, a study may use an AI classifier to assign engagement categories to student responses.
The research team should define the construct, classification framework, validation criteria, and procedures for uncertain cases.
Where the proposed measurement process affects the approved protocol or participant risks, additional review may be required.
Even when formal ethics approval is unnecessary, methodological validation remains essential.
Approval of a procedure does not independently establish that its measurements are valid.
What Should Be Submitted for an AI-Related Protocol Review?
Requirements vary, but a submission may need to explain the proposed AI activity and its implications for the study.
Relevant information can include the intended task, system or provider, categories of data processed, participant-facing interactions, safeguards, validation procedures, and changes to consent or research materials.
Researchers should also explain who will oversee consequential AI outputs and how errors will be handled.
The responsible review office determines the documentation required under its procedures.
Providing a clear account of the proposed use is generally more useful than simply stating that artificial intelligence will assist the research.
When Should Researchers Delay AI Use?
Researchers should delay a proposed AI activity when necessary authorization has not been established, when the processing arrangement cannot adequately protect restricted information, or when consequential outputs lack sufficient validation.
The appropriate response may involve selecting an institutionally approved service, modifying the processing procedure, obtaining additional review, or using a non-AI alternative.
Delay should not be interpreted as a permanent prohibition. A revised arrangement may become acceptable when the relevant requirements are satisfied.
However, retrospective approval should not be assumed to remedy an unauthorized disclosure or previously implemented protocol deviation.
