What Should Researchers Do if They Decide to Use AI After Ethics Approval Has Already Been Granted?

What Happens if You Want to Use AI After Your Research Has Already Been Approved?

Your qualitative research study has received ethics approval. You have recruited participants, conducted interviews, and begun preparing the transcripts for analysis.

When the protocol was submitted, you planned to analyze the interviews using conventional qualitative analysis software and researcher-led coding. Generative AI was not included in the approved methodology.

Now you discover that an AI tool could help summarize transcripts, identify possible patterns, and organize preliminary codes. You believe it could make the analytical process more manageable without changing your research questions.

However, the original approval was granted for a particular set of procedures. It did not necessarily consider the AI provider, the information that might be submitted, or the safeguards required for the new processing arrangement.

Can you simply introduce AI because the study is already approved, or must you obtain an institutional determination before changing the workflow?

Review the Proposed Change Before Implementing It

How to Introduce AI Into an Already Approved Research Study

Research Ethics Approval Applies to an Approved Set of Procedures

Research ethics approval is generally granted on the basis of the protocol and supporting materials submitted for review, subject to the applicable institutional conditions.

These materials may describe participant recruitment, informed consent, data collection, analysis, confidentiality safeguards, and anticipated risks.

Approval should not be interpreted as unrestricted permission to modify those procedures whenever a new technology becomes available.

For example, a protocol may authorize the research team to analyze interview transcripts stored on institutionally managed systems. Introducing an external AI service could create a different data processing arrangement.

The research questions might remain identical, but the information flow and confidentiality conditions may change.

Under the United States Common Rule, 45 CFR 46.108(a)(3)(iii) requires institutional review board procedures to ensure that changes in approved research are not initiated without review and approval, except when necessary to eliminate apparent immediate hazards to participants. This requirement applies within that regulatory framework and should not be treated as the governing rule for every institution internationally.

Researchers should follow the post-approval modification procedures applicable to their particular study.

First Determine What Has Actually Changed

The first step is to compare the approved procedure with the proposed AI-assisted activity.

For example, if AI was already authorized to assist with the analysis of appropriately protected transcripts under a specified service arrangement, routine use within those conditions may not constitute a new protocol change.

However, switching to a different provider, introducing identifiable information, or allowing AI to interact directly with participants may raise additional considerations.

The assessment should focus on the actual differences rather than assuming that every use of a new AI tool is a substantial protocol modification.

Distinguish a Minor Change From a Consequential Modification

Research institutions may distinguish administrative changes, minor modifications, and changes requiring more extensive review. The terminology and procedures vary.

For example, correcting a typographical error in a nonconfidential study document is unlikely to have the same ethical significance as introducing a new external processor for identifiable participant records.

However, researchers should not assume that they can classify every change independently. Some institutions require even minor modifications to be documented or submitted through a particular process.

These descriptions are practical distinctions, not formal institutional classifications. The applicable review body determines how the proposed modification should be handled.

Do Not Upload Participant Data While Waiting for an Ethics Determination

If the proposed AI activity may require additional authorization, researchers should avoid implementing it before the necessary determination has been obtained.

For example, a researcher should not upload identifiable transcripts to an unapproved external service merely to test whether the analysis produces useful results.

Even a preliminary upload may constitute the very processing activity that requires review.

Deleting the conversation afterward does not necessarily reverse the original transfer or establish that the provider retained no information.

Researchers can instead explore the tool using fictional examples, nonconfidential material, or other appropriately authorized test data while the institutional requirements are clarified.

The important distinction is between evaluating a tool's general capabilities and processing protected research information through that tool.

Review the Original Consent Materials

Introducing AI after approval may raise questions about whether the new processing arrangement is consistent with what participants were told when they agreed to participate.

Suppose participants were informed that their interviews would be analyzed by the university research team and stored on institutionally managed systems.

Submitting those interviews to an external AI provider may differ from the processing arrangement participants reasonably understood.

Whether additional information or renewed consent is required depends on the original consent language, applicable laws, institutional requirements, and the proposed change.

Researchers should not automatically assume that any mention of data analysis authorizes every possible analytical technology.

Equally, they should not assume that all AI-related modifications require participants to sign a new consent form. The reviewing institution should determine the appropriate procedure.

This is related to participant expectations about AI use, which may remain ethically important even where a processing arrangement appears technically permissible.

Examine Whether AI Introduces a New Data Recipient

One of the most consequential changes may concern where research information is processed and which organizations can access it.

For example, a researcher may originally have planned to store and analyze interview transcripts entirely within institutional systems.

Introducing a cloud-based generative AI service could involve an external provider, additional infrastructure, or different contractual conditions.

Relevant considerations include data retention, permitted uses, access controls, international transfers, deletion arrangements, and the provider's role under applicable data-protection law.

However, external processing is not automatically prohibited. Some institutions authorize specific AI environments under contractual and technical safeguards.

Researchers should verify whether the selected service is authorized for the particular category of research information.

The question of whether an AI provider becomes a recipient of research data should be assessed using the actual processing arrangement rather than the product name alone.

Assess Whether AI Changes the Original Risk-Benefit Balance

A new AI workflow may offer practical benefits, including faster organization of documents or assistance with preliminary analysis.

However, these benefits should be considered alongside any changes in confidentiality, analytical accuracy, participant welfare, or the consequences of incorrect outputs.

For example, an AI system may produce speculative classifications about participants' professional competence or emotional states even when those attributes were not part of the original study.

If such outputs are used in analysis or reporting, they may introduce risks not considered during the initial review.

Researchers should evaluate whether the proposed AI activity changes the likelihood, severity, or distribution of potential harms.

The relevant issue is whether AI changes the ethical risk of the study, not merely whether the new workflow is faster or more convenient.

What Should a Protocol Amendment Include?

When the institution determines that an amendment is required, the researcher should describe the proposed changes accurately and provide the information needed for review.

The precise form and supporting documents depend on institutional procedures.

Researchers should avoid inventing technical safeguards or asserting that the new system is compliant without confirming the applicable conditions.

The amendment should describe the proposed procedure, not simply announce that AI will be introduced.

How Should You Explain the Proposed Change to the Ethics Committee?

A useful explanation distinguishes the original approved workflow from the proposed AI-assisted alternative.

This example assumes that an institutionally authorized system is available. Researchers should not use that description unless the authorization and safeguards have actually been verified.

Where no suitable system has been approved, the proposed amendment should accurately describe that unresolved issue.

Can AI Help Prepare the Amendment Request?

Generative AI can assist in comparing protocol versions, organizing proposed changes, and drafting a clear description for institutional review.

However, the researcher must verify the content against the original approval and actual processing arrangements.

Researchers should avoid submitting confidential protocols or participant information to an unauthorized AI service while seeking assistance with the amendment.

Where necessary, use nonconfidential descriptions or an institutionally approved environment.

What If the Ethics Committee Determines That No Amendment Is Needed?

An institution may determine that a proposed modification does not require formal amendment approval.

For example, the change may fall within the original protocol or be classified as a minor administrative matter under the institution's procedures.

Researchers should retain documentation of the determination where appropriate and follow any conditions communicated by the reviewing office.

However, an ethics determination that no amendment is necessary does not automatically establish that every data-processing arrangement is authorized.

Separate information security, contractual, data protection, or institutional AI policies may still apply.

Researchers should therefore distinguish the absence of an amendment requirement from unrestricted permission to use a particular AI service.

What If the Researcher Has Already Used AI Without Checking?

Sometimes researchers may introduce AI before recognizing that the activity could require additional institutional review.

For example, a researcher may have uploaded participant transcripts to an external AI service without confirming whether the processing was authorized.

In such circumstances, the researcher should not assume that deleting the chat or removing the AI-generated outputs resolves the issue.

The appropriate response depends on what occurred, what information was involved, and the institution's reporting requirements.

Researchers should stop further unauthorized processing, preserve relevant factual records, and promptly consult the appropriate institutional office about possible protocol deviations, privacy incidents, or other reporting obligations.

Where personal data may have been exposed, the institution's data protection or information security personnel may also need to assess the incident.

Researchers should not conceal the activity or attempt to retroactively characterize it as approved.

Does a New AI Tool Always Require Another Amendment?

Not necessarily. The requirement depends on whether the new tool changes the approved procedure or processing conditions in a way that requires review.

For example, switching between tools within an already authorized institutional environment may be treated differently from transferring identifiable participant records to a new external provider.

Even when two AI systems perform the same analytical function, their data retention, access, security, and contractual arrangements may differ.

Researchers should evaluate the actual change rather than assume that all generative AI services are equivalent.

Where the institution has established a procedure for approving tools or processing environments, researchers should follow that procedure before changing systems.

What if Data Collection Is Already Complete?

Completing participant recruitment and data collection does not eliminate ethical responsibilities concerning the subsequent use of research information.

Data analysis, storage, sharing, retention, and reporting remain part of the research process.

For example, introducing AI during analysis may change where transcripts are processed or create new inferences about participants even though no further interviews will be conducted.

Researchers should therefore evaluate post-collection modifications according to their actual implications.

The fact that participants will not be contacted again does not automatically establish that additional consent, authorization, or review is unnecessary.

Can Researchers Use AI While the Amendment Is Pending?

Researchers should follow the institution's rules concerning implementation of proposed changes.

Where prior approval is required, the modified procedure should generally not begin until the appropriate authorization has been granted, subject to any applicable exceptions established by the governing framework.

Researchers may be able to continue activities already authorized under the original protocol, provided they remain within the approved conditions.

For example, a research team might continue its approved manual analysis while awaiting a determination about an external AI service.

However, researchers should not process participant information through the proposed new system merely because an amendment has been submitted.

Submission and approval are different stages of the institutional process.

Document the Final Authorized Workflow

After receiving the necessary determination, researchers should ensure that their actual procedures reflect the authorized arrangement.

Relevant documentation may include the approved modification, provider configuration, data categories processed, consent-related decisions, safeguards, and procedures for verifying AI-generated outputs.

If the final research report describes AI-assisted analysis, it should accurately distinguish what AI contributed from the analytical judgments made by the research team.

Researchers should also document consequential changes that occur later rather than assuming that one amendment authorizes every future expansion of AI use.

The broader obligation is to maintain consistency between the approved research procedures, actual conduct of the study, and the methodology ultimately reported.

Introducing Generative AI During Analysis of an Approved Qualitative Study

The study's research questions have not changed, but the proposed analytical procedure may involve a new data recipient and additional processing risks.

The appropriate response is therefore to evaluate the modification before implementation rather than assuming that the original ethics approval covers the new arrangement.

Common Mistakes When Introducing AI After Research Ethics Approval

Obtain the Appropriate Determination Before Changing the Approved Workflow

When considering AI after ethics approval, begin with the original protocol rather than the capabilities of the new tool.

Identify the precise modification, evaluate its implications for participant information and research procedures, and determine which institutional review or authorization process applies.

Retain documentation of the determination, approved conditions, and actual AI-assisted activities.

The objective is not to prevent reasonable methodological improvements after approval. It is to ensure that changes affecting the ethical conditions of the research are evaluated before participants or their information are exposed to unreviewed risks.

Before Introducing Generative AI After Ethics Approval

Frequently Asked Questions About Using AI After Ethics Approval

Existing Ethics Approval Does Not Automatically Authorize Every New AI Workflow

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.