U-M Research & Compliance

AI Use in Human Research Studies

Overview

This guidance outlines IRB expectations for the use of artificial intelligence (AI) in human research conducted at the University of Michigan. AI technologies may be used to collect, process, or analyze data or to generate outputs that inform study activities or participant interactions. This guidance does not pertain to the study team’s use of AI tools for project tasks, such as protocol development, meeting notes and summaries where identifiable data is not discussed, etc.

When used in human research, AI tools may raise considerations that differ from traditional research methods. These may include issues related to data privacy and confidentiality, accuracy and reliability of outputs, bias and data quality, and the level of transparency or human oversight.

For example, AI systems can generate inaccurate or fabricated results (i.e., “hallucinate”), re-identify data previously considered de-identified, or change in performance over time (i.e., reducing or changing the accuracy and reliability of outputs). Some AI platforms may rely on external or third-party services, which may introduce additional data security or participant protection concerns.

The use of AI in research may also affect the informed consent process, participant understanding, and autonomy, and may introduce risks that evolve throughout the study. Researchers must assess these risks, apply appropriate safeguards, and clearly describe how AI will be used in the eResearch application, research protocol, and consent materials.

Examples of AI Used in Human Research

The categories below describe some examples of AI used in human research. A study may use more than one type (e.g., a chatbot that also summarizes text). Knowing which type(s) of AI the study is using can help describe the study clearly, anticipate potential risks, and identify appropriate privacy, bias, and safeguards.

Generative AI (GenAI)

Chatbot, writing and summarizing tools, and study assistants

Predictive machine learning (ML)

Predicting outcomes or generating risk scores   

Rule-based systems (expert systems)

If/then logic for screening or workflows 

Computer vision

Analyzing medical images or video

Personalization/recommender systems

Tailored messages or recommendations

Agentic AI

Systems that perform multi-step actions using external tools

Natural language processing (NLP)

Analyzing study notes and narratives; extracting information from documents     

Hybrid / human-in-the-loop systems

AI outputs reviewed or confirmed by human experts

Deep learning (neural networks)

Advanced models for imaging or sensor data

Speech and audio AI

Speech-to-text or sound analysis

Causal Inference

Estimating the effects of an intervention (e.g., what works and for whom)

Adaptive decision systems

Systems that adjust over time based on feedback

eResearch Application

Beginning in May 2026, the IRB eResearch HUM application includes new questions to identify and describe the use of AI tools in research studies.

Identifying AI Use (Question 1)

Select “No use of AI technology, or use limited to supporting tasks” when AI is used only for general project support (e.g., drafting documents or meeting notes) and not for participant interaction, data collection, or analysis of research data.
Select a “Yes” response when AI is used in the conduct of the research, including interaction with participants, analysis of research data, or generation of outputs that inform study activities.

Identifying AI Tools (Question 1.2)

Indicate how the AI tool is provided:

  • U-M AI Tools – Tools supported by U-M Information and Technology Services (ITS).
  • Commercial or Public Tools – External or publicly available AI platforms (including U-M–licensed services)
  • Other – Tools developed by the research team or collaborators

Application Expectations

In addition to completing the AI questions, study teams should:

  • Describe how AI is used in the study
  • Identify the AI tools used
  • Describe data inputs and outputs, including any external or third-party AI tools or platforms
  • Address risks related to privacy, confidentiality, bias, and reliability

Information entered in other smartform sections (e.g., Benefits and Risks, Confidentiality, Security, and Privacy) should align with and support the description of AI use.

Protocol and Supporting Documentation

Additional detail should be included in the study protocol, when applicable, to provide a full description of AI use.

Projects that do not require a separately uploaded protocol (e.g., most secondary-use-only research or Exempt studies) should ensure that all relevant considerations are addressed within the application or may choose to include a protocol document (e.g., HRPP Exempt or Comprehensive templates).

Consultation and Review

When appropriate, the IRB may recommend consultation with U-M Information Assurance or other subject matter experts to ensure adequate oversight and protection of research participants. This may include consultation with Security Unit Liaisons for IT-related or data security considerations.

Study teams should include in the IRB application reference to any prior or ongoing consultations (e.g., Information Assurance, Clinical Intelligence Committee (CIC) (Level-2 login required), MIDAS).

Amendments

Changes to AI use in a study must be reviewed and approved by the IRB prior to implementation in accordance with standard Amendment requirements (IRBMED and IRB-HSBS).

U-M AI Tools

U-M provides AI tools supported by Information and Technology Services (ITS) that are designed to meet institutional data security and privacy requirements. Study teams may be required to use U-M–supported AI tools when working with research data, particularly when data are sensitive or identifiable.

Third-party Tools

If you plan to use a third-party AI tool, whether or not licensed by U-M, consult with U-M Information Assurance (ITS or HITS) as needed. 

Do not submit personally identifiable or sensitive information to third-party or public AI tools.  Such data may only be used with AI tools that are approved for the applicable data classification level and have appropriate data protection agreements in place.

Protocol Guidance

Protocols that include the use of AI in human research should provide sufficient detail to support IRB review of study risks, data protections, and procedures. Clear documentation supports IRB review and helps ensure that participants are adequately protected and that AI use is fully understood within the research context.

The following considerations can help study teams prepare protocols that provide sufficient information for IRB review of AI use:

AI System and Use

  • Type of AI system and its role (e.g., participant interaction, data collection, or data analysis)
  • Whether the AI system is static, adaptive, or operates with limited human oversight
  • Whether the AI tool is experimental or being developed as part of the study
  • Justification for the use of AI, including why AI is appropriate for the study objectives
  • The role of the AI outputs in final decision-making and how outputs will be evaluated
  • How AI use will be explained to participants, or a justification when disclosure is not appropriate

Data Sources and Data Flow

  • Characteristics and types of data used and generated by the AI system, including what data will be collected, entered, generated, or combined
  • Data collection methods (e.g., participant entry, study team entry, automated data collection, such as scraping)
  • A justification for the data used, including why each type of data is necessary for the study
  • Assurances that data collection is limited to the minimum necessary to achieve the study objectives
  • Whether data will be shared with external or third-party AI tools or platforms, including what data will be shared and under what conditions
  • Whether participant data will be used by the AI system, including for training or improvement

Privacy and Confidentiality Protections

  • Measures to protect participant privacy and data confidentiality
  • Risks related to re-identification or inference of sensitive information by the AI tool
  • Data security safeguards (e.g., access controls, encryption, or secure storage)

Data Quality, Bias, and Impact on Participants

AI systems may reflect limitations in the data used to develop or apply them. When applicable, protocols should address: 

  • Known limitations or potential sources of bias in datasets or AI models that may affect study results or participant outcomes
  • Steps to evaluate and manage potential bias, as appropriate to the study design and available data
  • Whether the AI system may perform differently across relevant populations or data sources, if this could affect the validity of results or risks to participants

Accuracy, Reliability, and Oversight

  • How AI use and outputs will be validated for the study population and purpose
  • Plans for monitoring AI performance, including potential model drift
  • The role of human oversight in reviewing or confirming AI outputs
  • Procedures for managing errors or unexpected results

Informed Consent Guidance

When AI is used in human research, consent materials must clearly describe, in lay language, how AI is involved and any associated risks to participants. The level of detail should reflect the significance of the AI’s role in the study.

The following considerations can help study teams prepare consent materials that provide clear and accurate information to participants:

Use of AI in the Study

  • Description of the AI system and its role (e.g., participant interaction, data collection, or analysis)
  • Whether the AI tool is experimental or being developed as part of the research
  • Whether AI outputs will inform or influence decisions affecting participants
  • Whether participant data will be used to train or improve AI systems

Risks of AI Use

  • Potential for inaccurate, incomplete, or misleading outputs (sometimes referred to as “hallucinations”)
  • Potential for bias or unequal impacts across populations
  • Privacy risks, including possible re-identification of data
  • Psychological, social, or other potential impacts related to AI use

Privacy and Confidentiality

  • What data will be collected and how it will be used, protected, stored, and shared
  • Whether data will be shared with third-party or external AI tools
  • Any limitations to confidentiality, particularly when using external systems

Data Use and Future Use

  • Whether data will be combined with other datasets for the current project and/or future use
  • Whether data will be retained after participant withdrawal
  • Whether data will be reused, shared, or commercialized, and if participants will share in any profit
  • Whether participants’ data can be withdrawn once entered into the AI system

Resources and References

U-M offers resources and guidance to support the appropriate and compliant use of AI in human research. These materials provide information on AI tools, data security, and training opportunities to help research teams responsibly integrate AI technologies into their studies.

U-M AI – Guidance and Training

AI Tools and Services

Data Security and Compliance

Other U-M Guidelines for Secure AI Use

Definitions and Reference Materials

Questions?

U-M Institutional Review Boards (IRBS)

IRB-Health Sciences and Behavioral Sciences (IRB-HSBS)
Phone:  734-936-0933
Email:  [email protected]
IRB-HSBS Website

IRBMED
Phone:  734-763-4768
Email:  [email protected]
IRBMED Website

Other questions:

If you have questions, concerns, or suggestions about human research protections at U-M but you’re not sure who to contact, email us at [email protected].