03 · What You Need to Know
Why Sensitive Research Information Requires More Than Automated Interpretation
What Makes Research Information Legally or Ethically Sensitive?
Research information may be sensitive because of its content, the circumstances under which it was collected, the identities of the people involved, or the consequences of disclosure or misinterpretation.
Examples include participant accounts of discrimination, allegations of misconduct, medical information, personal histories, legally protected records, and confidential institutional documents.
Some information may be legally protected under applicable privacy or confidentiality requirements. Other information may be ethically sensitive even when it does not fall within a particular statutory category.
For example, a participant's description of an embarrassing workplace incident may not automatically qualify as sensitive personal data under every legal framework, but unauthorized disclosure could still cause reputational or professional harm.
Researchers should therefore evaluate both legal classifications and contextual ethical concerns.
Sensitivity is not determined solely by the presence of obvious identifiers.
Legal Interpretation and Ethical Interpretation Are Different
Legal interpretation concerns the meaning and application of laws, regulations, contractual obligations, and other legally relevant authorities.
Ethical interpretation concerns questions about appropriate conduct, participant welfare, fairness, respect, confidentiality, and the responsibilities associated with research.
The two may overlap, but they are not interchangeable.
Legal interpretation
Determining how applicable legal authorities and obligations relate to a particular situation, with attention to jurisdiction and relevant facts.
Ethical interpretation
Evaluating responsibilities, competing interests, possible harms, and appropriate conduct using relevant ethical principles and research context.
An activity may comply with a particular legal requirement while still raising ethical concerns.
Conversely, a researcher may believe an action is ethically justified while remaining subject to legal restrictions.
AI-generated explanations should not collapse these distinct questions into a single recommendation.
Why Generative AI Can Produce Incorrect Legal Interpretations
Generative AI systems can produce plausible legal explanations without reliably establishing that the cited authorities exist or apply to the situation.
A system may invent judicial decisions, misstate statutory provisions, overlook amendments, or apply rules from the wrong jurisdiction.
These problems are especially consequential when researchers use AI to determine disclosure obligations, participant rights, contractual restrictions, or procedures for handling sensitive information.
For example, an AI assistant might incorrectly state that a researcher is legally required to disclose an interview transcript to a particular authority.
Acting on that recommendation without verification could compromise confidentiality or create other consequences.
Researchers should verify legal claims using authoritative sources and obtain qualified legal guidance where interpretation is consequential.
A confident AI response does not establish legal accuracy.
Why Jurisdiction Matters
Research collaborations may involve participants, investigators, institutions, and data-processing services located in different jurisdictions.
Legal obligations can vary according to the applicable laws, institutional arrangements, and circumstances of the research.
For example, personal data processing may be subject to the Philippine Data Privacy Act of 2012, the European Union's General Data Protection Regulation, or other relevant frameworks depending on the situation.
Researchers should not assume that one jurisdiction's rules automatically apply to every international collaboration.
Nor should they assume that the location of an AI provider alone determines the complete legal framework.
Questions concerning territorial scope, contractual obligations, lawful processing, and international transfers may require specialist assessment.
AI can help researchers identify possible issues for investigation, but it should not independently determine which laws govern a complex research arrangement.
Why Ethical Interpretation Requires Context
Ethical questions frequently involve competing considerations that cannot be resolved through a generic rule.
For example, researchers may need to balance participant confidentiality against concerns about potential harm, institutional reporting obligations, or the expectations established during informed consent.
The appropriate response may depend on the research protocol, the nature of the information, the seriousness of the concern, and applicable institutional procedures.
An AI system may generate an apparently decisive recommendation while overlooking these contextual factors.
Researchers should therefore distinguish between using AI to organize ethical questions and allowing it to determine the final ethical response.
Qualified research ethics personnel, institutional review bodies, and relevant specialists may be necessary when decisions exceed the researcher's expertise or authority.
Informed Consent Does Not Automatically Authorize Every AI Use
Participant consent is relevant to determining how research information may be collected and used.
However, consent to participate in a study does not automatically establish permission to submit identifiable information to an external generative AI service.
The proposed processing must be evaluated against the actual consent arrangements, approved protocol, applicable data-protection requirements, and institutional obligations.
Researchers should examine whether the AI processing is consistent with the purposes communicated to participants and whether additional authorization is necessary.
Where consent is not the applicable legal basis for processing, other legal and ethical requirements may still govern the activity.
AI-generated interpretations of consent language should not replace institutional or legal review when the consequences are substantial.
Why Deidentification May Not Eliminate Sensitivity
Removing names and direct identifiers can reduce some privacy risks, but it does not necessarily make research information anonymous.
A participant may be identifiable through a distinctive combination of circumstances, institutional position, location, or reported experiences.
For example, an interview transcript describing a unique administrative role and a widely known workplace incident may reveal the participant's identity even without a name.
Researchers should assess reidentification risks in context.
They should also distinguish information that has been deidentified from information that meets the applicable legal standard for anonymization.
These distinctions matter when determining whether sensitive material can be processed through an AI service.
AI Can Misinterpret Participant Meaning
Qualitative research often involves interpreting language within social, cultural, and institutional contexts.
A participant's account may include irony, hesitation, indirect criticism, culturally specific expressions, or references that require contextual understanding.
Generative AI may interpret such statements differently from the participant's intended meaning.
For example, a participant describing institutional pressure may use figurative language that an AI system interprets as a literal threat.
Alternatively, an indirect description of discrimination may be overlooked because the system expects explicit terminology.
Researchers should not treat AI-generated interpretations as direct evidence of participants' intentions, experiences, or psychological states.
Interpretive claims should remain grounded in the research material and appropriate methodological procedures.
Why Allegations Must Not Be Treated as Established Facts
Sensitive research information may contain allegations concerning identifiable individuals or organizations.
An interview participant may describe conduct they believe was discriminatory, unlawful, or unethical.
The researcher's responsibility is to represent the account accurately within the approved research methodology.
An AI system may inadvertently convert an allegation into a statement of established wrongdoing.
For example, a generated summary might describe an institution as having violated a law when the source material only records a participant's perception of an incident.
Such transformations can create serious ethical and reputational concerns.
Researchers should distinguish reported experiences, interpretations, corroborated evidence, and legally established findings.
AI-generated language should be checked carefully when it concerns allegations or potentially stigmatizing claims.
What Happens When Ethical Principles Conflict?
Research ethics decisions may involve competing obligations.
For example, confidentiality may be important for protecting a participant's trust and privacy, while another concern may raise questions about foreseeable harm.
There is no universally applicable AI-generated answer that resolves every such conflict.
The appropriate response depends on the facts, applicable legal requirements, approved protocol, and institutional responsibilities.
Researchers should identify the relevant obligations and consult authorized personnel when necessary.
AI may help formulate questions for review, but it should not be allowed to invent an exception to confidentiality or authorize a disclosure.
When Can AI Be Used for Preliminary Organization?
Some AI-assisted activities may be appropriate when they do not require the system to make consequential legal or ethical judgments.
For example, researchers may use an authorized tool to organize nonconfidential policy documents or identify sections of an institutional guideline relevant to a question.
AI may also suggest a checklist of issues to discuss with a research ethics office.
However, even preliminary assistance requires attention to source accuracy and processing authorization.
| AI-assisted activity |
Possible appropriate role |
Necessary caution |
| Summarizing public research ethics guidance |
Preliminary orientation |
Verify against the current authoritative document |
| Identifying potentially relevant legal provisions |
Generating questions for legal review |
Confirm existence, applicability, and current wording |
| Organizing confidential interview material |
Restricted assistance where authorized |
Assess confidentiality, privacy, and interpretive risks |
| Interpreting participant allegations |
Suggesting possible analytical questions |
Avoid treating allegations or generated interpretations as established facts |
| Determining disclosure obligations |
Identifying issues for qualified consultation |
Do not rely on AI as final legal authority |
| Deciding how to respond to participant risk |
Support only within an approved and appropriately validated workflow |
Preserve authorized human judgment and escalation procedures |
The appropriate classification depends on the research context. The table does not authorize any particular AI service to process sensitive information.
Why Legal and Ethical Sources Must Be Verified
Researchers should distinguish between a genuine authoritative source and an AI-generated description of that source.
For legal questions, relevant primary materials may include statutes, regulations, official decisions, and applicable contractual provisions.
For research ethics questions, relevant materials may include approved protocols, institutional policies, professional guidelines, and authoritative ethical frameworks.
The World Medical Association's Declaration of Helsinki and ALLEA's European Code of Conduct for Research Integrity provide important principles within their respective domains.
However, general ethical guidance does not automatically resolve a particular legal obligation or replace institutional approval.
Researchers should examine the actual text, scope, and applicability of relevant authorities.
Why AI Interpretations Can Create Evidence Laundering
Evidence laundering can occur when an unsupported interpretation acquires apparent credibility because it is repeatedly presented as though it were established evidence.
For example, an AI system might incorrectly characterize a participant's statement as evidence of unlawful conduct.
If the researcher incorporates that characterization into an analytical memo, later summaries may treat it as an established finding.
Over successive stages, the original distinction between participant account and researcher interpretation may become obscured.
Researchers should preserve the provenance of sensitive claims and distinguish source material from AI-generated interpretation.
This is especially important when research outputs may affect identifiable people or institutions.
How Should Researchers Handle Potential Reporting Obligations?
Some research situations may raise questions about legally or institutionally required reporting.
However, the existence, scope, and exceptions to reporting obligations depend on the applicable jurisdiction and circumstances.
Researchers should not rely on a general-purpose AI assistant to determine whether confidential information must be disclosed.
Instead, they should consult the approved protocol, relevant institutional procedures, and qualified legal or professional personnel.
Where an immediate safety concern exists, researchers should follow established emergency and safeguarding procedures rather than delay action while seeking AI-generated advice.
The relevant response must be grounded in authorized responsibilities.
Why Human Oversight Must Include Appropriate Expertise
A researcher may understand the substantive research topic while lacking expertise in privacy law, clinical ethics, or institutional reporting obligations.
Reviewing an AI-generated interpretation without the necessary expertise may provide limited protection against error.
Meaningful oversight requires identifying who is qualified to evaluate the relevant issue.
For example, a data-protection officer may be needed to assess a proposed processing arrangement, while a research ethics committee may need to review a change affecting participant protection.
These responsibilities should not be assigned solely according to convenience.
The broader requirement for genuine human oversight applies particularly strongly when interpretations could produce serious consequences.
When Should AI Be Avoided?
AI should be restricted or avoided when sensitive information cannot be processed through an authorized environment, when the system's interpretation cannot be adequately verified, or when errors could cause serious harm.
It may also be inappropriate when the task requires confidential legal advice, specialized ethical judgment, or an official institutional determination that the AI system is not authorized to provide.
Researchers should distinguish between using AI to prepare questions for an expert and allowing AI to replace that expert's judgment.
Where the necessary safeguards are unavailable, the appropriate decision may be to use established human-led procedures.
This reflects the principles of risk-proportionate AI use in high-stakes research.
