Does Informed Consent Need to Mention AI Processing of Participant Data?

Do Participants Need to Be Told That AI Will Process Their Data?

You plan to use AI to transcribe interviews, classify survey responses, analyze images, extract information from clinical records, or assist with another research task. Should the consent form explicitly tell participants that artificial intelligence will process their data?

There is no universal rule requiring every consent form to contain the words “artificial intelligence.” Research studies routinely use software, statistical packages, transcription systems, cloud services, and other technologies without listing every technical component in participant-facing materials.

But AI can become ethically significant when it changes something participants would reasonably need to understand before deciding whether to participate: where their information goes, who processes it, what information may be inferred, whether an external provider is involved, or what risks arise from the processing.

The real question is therefore not simply whether AI appears somewhere in the workflow. It is whether the AI use is material to informed participation.

Mention Material AI Processing, Not AI Merely for the Sake of Naming It

What Participants Actually Need to Know About AI Processing

Informed Consent Is About Meaningful Information, Not a Technology Inventory

A research consent form is not normally a complete technical specification. Listing every operating system, statistical library, database engine, and software version would make consent longer without necessarily making it more informed.

The same principle can apply to AI. If an AI component functions as an internal analytical tool and does not materially alter participant risk, data recipients, confidentiality arrangements, or the nature of participation, naming the technology may add little.

The answer changes when the technology affects information that a reasonable participant would likely consider relevant to the decision to participate.

For example, “your interview will be analyzed by the research team” conveys something different from a workflow in which identifiable recordings are transmitted to an external commercial AI service for transcription and analysis. The difference is not that the second workflow contains fashionable software. The difference is what happens to participant information.

AI May Need Disclosure When It Changes Who Processes the Data

If participant information will leave the research team's controlled environment and be transmitted to an external AI provider, that may be material to confidentiality and privacy.

The precise legal characterization of the provider varies among jurisdictions and contractual arrangements. An AI provider may function as a processor, service provider, business associate in certain U.S. health contexts, or another type of recipient.

Researchers should therefore determine whether AI processing amounts to sharing participant data with a third party under the framework governing the study. If it does, the consent and privacy materials may need to explain that arrangement at an appropriate level.

AI May Need Disclosure When It Creates New Types of Inference

Some AI systems do more than automate a task researchers could readily perform manually. They may classify participants, predict outcomes, derive latent characteristics, detect patterns across datasets, or generate inferences about information participants did not explicitly provide.

SACHRP has highlighted this concern in human-subjects research involving AI, noting that AI and machine-learning systems may infer information that individuals did not knowingly disclose and that large-scale data combination can create privacy and group harms not adequately captured by traditional assumptions about identifiability.

If such inference is central to the study, participants may need to understand what kinds of conclusions will be drawn from their information, particularly where the resulting classifications are sensitive or consequential.

AI May Need Disclosure When It Changes Privacy or Confidentiality Risk

Consider two transcription workflows. In the first, a trained member of the research team transcribes interviews within an approved institutional environment. In the second, identifiable audio files are uploaded to an external AI transcription service.

Both produce transcripts. They do not necessarily create the same data-processing arrangement.

For research involving recordings, images, clinical information, detailed narratives, or other identifiable material, researchers should consider whether AI processing introduces a disclosure or retention pathway that participants would reasonably expect to know about.

This is particularly relevant when participant audio or video recordings are processed by AI, because the identifiable source material must generally reach the processing system before any de-identified transcript or derivative output can be produced.

Consent to Participate and Data-Protection Consent Are Not Always the Same Thing

Researchers should be careful with the word consent. Ethical consent to participate in research and consent as a legal basis for processing personal data can be distinct concepts.

The UK Information Commissioner's Office, for example, explicitly distinguishes consent to participate in a research study from consent used as the lawful basis for personal-data processing under the UK GDPR. A research project may ethically require participant consent while relying on another lawful basis for processing personal data.

This distinction matters because saying “the participants consented” does not, by itself, answer every data-protection question surrounding AI processing.

Researchers need to establish which ethical, regulatory, and legal requirements apply rather than treating one signature on a consent form as universal authorization for all data processing.

Transparency Can Matter Even When Consent Is Not the Legal Basis

Even where an institution relies on a lawful basis other than data-protection consent, transparency obligations and research-ethics expectations may still require participants to receive meaningful information about how their data will be used.

The ICO's current research guidance emphasizes that research-related processing must still be fair and lawful and must employ appropriate safeguards. Its guidance also notes that consent to research participation should not be confused with the separate question of the lawful basis for data processing.

Researchers should therefore avoid the false choice between “obtain consent for AI” and “say nothing about AI.” Depending on the framework, transparency about processing may be appropriate even where consent is not the legal mechanism authorizing that processing.

Consent Should Explain What Matters About AI, Not Merely Use the Acronym

A statement such as “AI will be used in this study” is technically transparent but often practically uninformative.

Where AI processing is material, participants may need information such as:

• what participant information will be processed;
• what the AI system will be used to do;
• whether identifiable information will be sent outside the research institution;
• whether an external service provider will process the information;
• what important confidentiality or privacy risks arise;
• whether AI will generate classifications, predictions, or other consequential inferences; and
• what safeguards will be used.

The level of detail should be proportionate. Participants generally do not need a lecture on transformer architecture to decide whether an interview recording may be sent to an external transcription service. Consent committees everywhere may quietly thank you for resisting the temptation.

Brand Names May or May Not Belong in the Consent Form

Whether to name a specific provider such as OpenAI, Microsoft, Google, Anthropic, or another company is a separate question from whether AI processing should be disclosed.

A brand name may be useful when the identity of the provider materially affects participant understanding or is required by institutional policy. In other situations, describing the category of processing and relevant data-sharing arrangement may be clearer and more durable, especially if an approved provider could change during a multi-year study.

Do not deliberately obscure an important recipient behind vague language merely to preserve flexibility. Equally, avoid making the consent form dependent on a product name when the ethically relevant fact is the controlled use of an approved external processor under specified conditions.

AI Used Only on Non-Participant Material Is a Different Question

Suppose a researcher uses generative AI to improve the wording of recruitment materials, develop a coding template using fictional examples, troubleshoot statistical code, or summarize public methodological literature. Those activities may involve no participant data at all.

There is generally little reason to insert such uses into participant consent merely because AI appears somewhere in the research workflow, unless an applicable institutional policy specifically requires it.

The consent question becomes more consequential when AI directly processes information obtained from or about participants.

More Sensitive Data Usually Justify More Explicit Transparency

The materiality of AI processing increases when the underlying data are particularly sensitive or identifying. Clinical records, genomic information, identifiable images, recordings, trauma narratives, illegal activity, sexual information, political information, or data concerning vulnerable populations may justify greater specificity.

For example, clinical research data processed with AI may involve overlapping health-information, privacy, research, contractual, and institutional requirements that make generic consent language inadequate.

The Ethics Committee Should See the Same Material Facts Participants See

If AI processing is sufficiently important to affect participant risk, confidentiality, data recipients, or study procedures, it should also be accurately described in the protocol submitted for ethics review.

The consent form and protocol should not tell two different stories about how participant information will be handled.

If the study is already underway and AI was not part of the approved workflow, the issue shifts from prospective disclosure to whether an ethics amendment is required before introducing AI.

When AI Processing Is Material Enough to Disclose

Common Mistakes About AI and Informed Consent

Ask Whether AI Changes Anything Participants Would Reasonably Want to Know

When designing consent materials, focus on materiality rather than the novelty of the technology. The more AI changes the participant-data relationship, the stronger the case for explicit disclosure.

For a new study, this assessment can be built into the research design from the beginning. For an existing study, the harder question is whether consent already obtained from participants covers AI use that was not anticipated when the study began.

Before Finalizing Consent for a Study That Uses AI

Questions About Informed Consent and AI Processing

Consent Should Explain Material AI Processing, Not Merely Name AI

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.