Considerations for the Responsible and Ethical Use of AI

• Ask if AI-generated ideas reflect Ut Prosim and the goal of improving the human condition.
• Pursue AI-inspired ideas for scholarly or societal merit, not solely for novelty or convenience.

• Use AI to explore existing research and refine early ideas.
• Transparently document how AI contributed to idea generation.
• Reflect on potential harms or unintended consequences that could affect individuals, groups, or systems beyond the research team, including societal, environmental, or ethical impacts, before pursuing AI-generated ideas.

• Treat AI suggestions as inputs for brainstorming, not as research decisions.
• Avoid relying on AI to define your research problem.

• Ask early questions about possible ethical or societal impacts before deciding to pursue the idea.
• Use caution when advancing AI-generated directions that could pose ethical or dual-use risks (research that has legitimate scientific value but could be misapplied to cause harm or create security vulnerabilities).

• Keep notes on when AI contributed to conceptualization.
• Follow University Libraries’ Guidelines for Citing AI
• Do not present AI-generated hypotheses as your own intellectual output.

• Archive AI prompts or outlines in research notebooks.
• Make final decisions about study focus and goals as a research team.
• Independently verify the accuracy of any AI-generated facts, claims, or citations. Treat all AI outputs as provisional until confirmed as tools can introduce errors, omissions, or fabricated references.

• Use only Virginia Tech-approved AI tools.
• Do not enter confidential or proprietary information (including grant concepts, unpublished data, inventions, or other information that cannot be shared publicly) into tools that are not Virginia Tech-approved.

 AI Tools and Access at Virginia Tech | Artificial Intelligence | Virginia Tech

• Combine AI-assisted searches with traditional internet, digital library, and database methods to ensure completeness.
• Avoid relying solely on AI summaries or rankings/ratings to determine relevance.
• If you are conducting a systematic review or other formal evidence synthesis research, consult guidance or get assistance in planning use of AI within your process.

• Use AI to identify underexplored topics or cross-disciplinary trends.
• Verify that AI-suggested sources and topics align with your research goals and ethical standards.

• Use AI to organize valid references, extract information, prepare draft summaries, and identify themes, not to replace critical reading and evaluation.
• Always read and assess primary sources yourself.

• Disclose AI assistance in your methods or acknowledgments section when appropriate.
• Be aware of potential bias in AI literature discovery tools. For example, many overrepresent English-language content or publications from high-impact journals.

• Verify all AI-generated references and claims against original sources as AI tools can introduce errors, omissions, or fabricated references.
• Educate yourself regarding the date range and sources used by the AI tool to understand its limitations and biases.
• Do not cite unverified or AI-generated materials as primary evidence.
• Be aware that AI tools might have limited access to recent or specialized publications. AI tools can overemphasize highly cited or mainstream publications, reinforcing existing perspectives and missing emerging or diverse scholarship.
• Be aware that a valid link or real citation does not guarantee that the article supports the point the AI suggests. You must read the source and confirm that its findings, context, and conclusions align with your interpretation.

• Use only Virginia Tech–approved AI systems for managing literature and notes.
• Avoid using personal or public accounts for collaborative writing or AI-assisted editing.

• Use AI to streamline simulations or optimize workflows that support scientific advancement.
• Apply AI to improve research design, transparency, and collaboration in ways that advance knowledge, serve the public good, and uphold Ut Prosim.
• Do not use methods that prioritize efficiency over accuracy or research quality.

• Experiment with AI-driven modeling or simulation tools that enhance discovery.
• Reflect on how new AI-assisted methods could create inequities, environmental burdens, or unintended ethical consequences.

• Evaluate whether AI assistance improves or hinders understanding and accessibility.
• Use AI to explore new analytical directions or generate design ideas that expand your team’s capability.
• Avoid using AI in ways that deskill researchers, obscure human contribution, or prevent students and trainees from developing the methodological expertise essential to their disciplines. Responsible AI use should enhance, not replace, mentorship and skill-building.

• Consider safety, consent, and compliance implications early, especially for studies involving human data or interventions.
• Carefully consider and document potential dual-use applications (ideas with both beneficial and harmful potential). If dual-use applications exist, use extreme caution when using (especially uploading) any research data. If research is funded, consult with your project/program officer or funding entity for guidance on AI use.

• Include a short statement in your methods section noting where AI contributed to design or parameter selection.
• Review available documentation from the AI tool to understand its training data sources, known limitations, and potential bias.
• Assess whether the AI’s outputs might systematically favor or exclude certain populations, variables, or perspectives relevant to your field.
• When developing or sharing data that could contribute to future AI training, document the dataset’s origins, composition, and intended use. Clearly describe potential limitations or biases (for example, demographic underrepresentation) and include metadata that guides responsible reuse. This transparency supports fairness, reproducibility, and ethical downstream AI development.
• Do not present AI-generated code or models as fully original.
• Do not rely solely on AI-generated methods or models without considering and reporting transparently how bias in the training data could affect validity, generalizability, or equity in your results.

• Independently verify AI-generated results whenever possible, and ensure that analytical steps are transparent and well documented. If human verification is not feasible, ensure transparency by documenting the limitations and describe how you validated the results.
• Review AI-generated analyses with disciplinary experts before adoption.
• Document model parameters and data inputs so results can be reviewed or replicated.
• Eligibility criteria for research participation must be applied in accordance with the IRB (Institutional Review Board)-approved study protocol. Automated screening tools (such as rule-based surveys) may be used when approved by the IRB. However, AI systems should not independently determine or infer participant eligibility, as a qualified researcher must retain oversight of inclusion and exclusion criteria.

• Use AI platforms that have been approved by Virginia Tech to collect, process, and store data at the appropriate risk level. For example, the Virginia Tech Advanced Research Computing (ARC) environment includes multiple clusters, each approved for specific data types (e.g., Controlled Unclassified Information (CUI), Moderate Risk data, and one for High-Risk identifiable data from humans).
• Never input regulated, confidential, or export-controlled data into tools that have not been approved to process them.

• Apply AI responsibly to improve efficiency, transparency, and reproducibility in research computing.
• Use AI-driven automation to expand accessibility or reduce human error in complex workflows.
• Avoid automation that prioritizes convenience over data integrity or research quality.
• Ensure AI-generated code, models, or outputs align with Virginia Tech’s Principles of Community by avoiding bias, exclusionary logic, or disrespectful content that could harm individuals or groups.

• Use AI to prototype tools that enhance human capability or scholarly insight.
• Use existing machine learning tools responsibly to support automation or analysis, and regularly assess tool performance, potential bias, and interpretability when applying them to your research.
• Consider whether automated systems could unintentionally replace critical human judgment, amplify bias, or create inequities in access, decision-making, or impacts.

• Use AI-generated code or machine learning-driven processes as learning and mentoring tools for students and staff developers.
• Encourage review and discussion of AI-generated code as part of team learning and peer mentoring.
• When creating software or automation that relies on AI tools, ensure the resulting outputs and workflows remain understandable, maintainable, and usable by humans. Design these workflows to expand human capability rather than creating systems only the AI or a few experts can operate.
• Avoid using AI in ways that prevent students and trainees from developing essential programming or analytical skills.

• Use AI tools approved by Virginia Tech that align with the data classification level.
• Attribute AI-assisted code generation transparently in documentation and publications.
• Do not deploy unverified AI-generated code in systems, especially with restricted data.

• Document when and how AI designed, generated, optimized, or tested code.
• When AI tools for code generation algorithm design, or workflow automation, review available documentation (e.g., known limitations, design assumptions, an potential bias) to understand how the tool may influence the logic or behavior of resulting software.
• Review documentation provided by AI tools to understand how their design or training might shape automated decisions or outputs.
• If you are developing or sharing data that might be used to train AI systems, document the dataset’s scope, sources, and known limitations. Include metadata and provenance that describe potential bias and appropriate contexts for reuse. Established frameworks such as Datasheets for Datasets and The Data Nutrition Project can be consulted for additional guidance.

• Review, test, and validate AI-generated code, machine learning outputs, and automated processes before implementation.
• Maintain version control and clear records of code and model review and approval.
• When using machine learning algorithms in automated workflows, maintain human oversight and verify results before integration into research conclusions.
• Do not attribute authorship to AI or rely on unverified AI-generated results or machine learning-driven automation for critical research functions. When verification is not possible, researchers should acknowledge this constraint and apply appropriate safeguards, such as transparency about uncertainty, expert judgment, and alternative validation approaches, to maintain research integrity.

• Use only Virginia Tech–approved AI and computing platforms for coding or automation that handle sensitive data.
• Follow Virginia Tech Risk Classification Standard when processing data with automated scripts, trained models, or AI-generated workflows.
• Never input or upload proprietary, confidential, or export-controlled data into publicly accessible code-generation tools.

• Apply AI to improve data accuracy, transparency, and reproducibility.
• Use AI to strengthen research that addresses community needs and societal challenges.
• Do not use data practices that prioritize convenience or automation over integrity, privacy, or inclusion.

• Use AI in ways that make data management efficient and reproducible while ensuring that resources and datasets are used fairly and responsibly.
• Consider how automation or model design might generate false conclusions, amplify bias, or exclude underrepresented populations.

• Use AI tools to support analytical thinking, not replace it.
• Encourage students and research staff to learn data literacy skills, even when AI automates parts of the process.
• Avoid overreliance on AI outputs without human review of assumptions, patterns, and limitations.

• Confirm that data processing tools and storage systems are approved for your data’s classification level.
• Verify IRB, FERPA, and export control compliance before sharing or analyzing data with AI systems.
• Document all data transformations performed by AI to ensure reproducibility and accountability.
• Avoid uploading sensitive, identifiable, or proprietary data to unapproved AI platforms.

• Examine whether AI tools or models might overrepresent dominant data sources or underrepresent marginalized groups.
• Document data provenance, preprocessing methods, and known limitations of AI analysis.
• If creating or sharing datasets that might be used in AI or machine-learning applications, include metadata describing potential bias, representational gaps, and appropriate reuse conditions. Tools such as the Data Nutrition Project Label Maker can support this process.
• Be transparent in publications or reports about where and how AI contributed to analysis or visualization.

• Independently validate AI-generated results through replication or secondary analysis.
• Maintain clear documentation of human oversight, quality control, and review processes.
• Do not rely solely on AI systems for analytical accuracy or discovery; humans must confirm findings and interpretations.

• Process data using Virginia Tech–approved tools matched to Virginia Tech’s data risk classification.
• Use ARC clusters or other university-managed platforms appropriate to the dataset (e.g., CUI, moderate risk, or high-risk data).
• Do not upload regulated or high-risk data into public or non-Virginia Tech-approved AI tools.
• Consult the Privacy and Research Data Protection Program (prdp@vt.edu) or the Office of Export and Secure Research Compliance (oesrc@vt.edu) if uncertain about approved systems or data classifications.

• Use AI to improve clarity and accessibility in scholarly communication, ensuring your work remains accurate and contributes meaningfully to your field.
• Review AI-generated outreach materials or summaries to ensure tone and inclusivity align with Virginia Tech’s Principles of Community.
• Collaborate with University Relations when using AI to draft press releases or public communications.
• Do not use AI to create or publish content that misrepresents research results or fails to meet university communication standards.

• Use AI to improve grammar, readability, or translation while retaining human authorship and intellectual control.
• Apply AI to adapt tone and format for audiences such as community partners or policymakers.
• Do not rely on AI for factual accuracy or citation generation; verify all results manually.
• Avoid allowing automation to oversimplify complex or nuanced findings.

• Retain full authorship of ideas, conclusions, and arguments; AI can assist with phrasing but should not entirely replace human reasoning.
• Encourage students and trainees to build writing and communication skills rather than outsourcing writing to AI.
• Use AI to improve accessibility, such as simplifying technical text for outreach audiences.
• Review and edit AI-generated text to ensure it reflects your reasoning, disciplinary vocabulary, and writing style. Remove or rewrite any sections that sound generic or disconnected from your research interpretation.

• Acknowledge AI assistance Using (and Citing) AI - AI: Tips & Tools from Your Librarians
• Verify that AI use complies with sponsor, contractual, and institutional policies.
• Never upload confidential, proprietary, or export-controlled content into tools that have not been approved by Virginia Tech to process restricted information.

• Follow the citation and acknowledgment examples provided by Virginia Tech Libraries for citing generative AI in academic writing.
• Consider using the GAIDeT Declaration Generator Tool to generate a clear disclosure of AI use across the research lifecycle.
• Include an acknowledgment in publications explaining AI’s role in text generation, summarization, or translation.
• Avoid minimizing AI’s contribution as being transparent strengthens credibility and ethical integrity.

• Review, fact-check, and edit all AI-generated text for accuracy, logic, and tone.
• Ensure conclusions and interpretations are written by humans, supported by evidence, and confirmed by human reasoning to the extent possible.
• Confirm translations and summaries are accurate and retain the intended meaning and context.
• Do not rely on AI-generated text for analytical reasoning, conclusions, or literature interpretation without checking and editing.

• Use only Virginia Tech–approved AI tools and secure platforms for writing and editing sensitive materials.
• Do not enter or upload proprietary, confidential, or fundable concepts as prompts in non-Virginia-Tech–approved AI tools. Once entered, that information is stored, shared with external servers, and might be used to train future models, potentially compromising intellectual property protection and your research ideas.
• Ensure translations and reformatting are performed in compliant environments when handling restricted or identifiable information.
• Consult the Privacy and Research Data Protection Program (prdp@vt.edu) and if uncertain about approved AI use for your research materials.

• Use AI to help identify areas of ethical risk, bias, or confidentiality concern that require further human review.
• Ensure AI-assisted ethical analysis aligns with the Principles of Community and policies on research integrity.
• When preparing ethics or compliance materials, consult any discipline-specific AI standards issued by your professional society to ensure your practices align with field expectations.
• Avoid using AI outputs as justification for ethical or regulatory decisions that have not been reviewed or approved by a qualified committee.

• Use AI to generate summaries of risk considerations, organize supporting documents, or help identify potential ethical concerns.
• Treat AI outputs as organizational tools; they can flag areas for attention but cannot evaluate ethical sufficiency.
• Avoid using AI to determine risk level or review classification. Only reviewers or committees trained in the applicable regulatory requirements can make those determinations.

• Use AI to assist with tasks such as bias detection or flagging sensitive data, while retaining human interpretation and approval.
• Ensure all human subjects, animal, or biosafety protections are confirmed by qualified personnel.
• Use AI for preparation and awareness, not compliance determinations.
• Do not rely on AI systems to determine whether research meets regulatory requirements or qualifies for exemption.
• AI cannot decide whether a study requires review or approval by the IRB, IACUC, or IBC. Researchers must consult the appropriate compliance office.

• Confirm that any AI-assisted review or documentation tools are approved for use with your project’s data classification level.
• Verify the accuracy and completeness of all AI-generated content before including it in ethics or compliance submissions.
• Do not input confidential protocols, reviewer comments, or proprietary materials into unapproved AI systems.
• Record how AI was used for transparency and reproducibility in ethics documentation.

• Treat AI-assisted outputs (e.g., risk summaries, bias flags, draft compliance text) as advisory and subject to human review.
• Ensure that all ethical determinations are reviewed and confirmed by appropriate institutional authorities.
• Maintain documentation of human decisions and approvals to support accountability and auditability.

• Use AI tools to identify potential bias or discrimination risks in study design, recruitment, or data interpretation.
• Disclose any AI assistance in ethics documentation or risk assessments submitted for review.
• Review all AI results manually; bias detection tools can misinterpret or exaggerate patterns if not validated by researchers.

• Use Virginia Tech-approved AI tools to assist with data confidentiality monitoring, such as detecting identifiable or sensitive information in datasets, text, or figures before public release.
• Periodically confirm that AI tools and storage systems meet Virginia Tech’s data-handling and confidentiality requirements.
• Document any AI-assisted confidentiality checks (e.g., dataset scans or content reviews) to support ethical transparency.
• Never rely solely on AI to ensure data protection—researchers retain responsibility for verifying the content of datasets.
• Do not upload confidential, restricted, or unpublished data into public AI systems.

• Use AI for grammar, structure, or formatting assistance in cover letters or submissions while maintaining your own scholarly authorship. 
• When creating public summaries or outreach materials, review AI-generated text to ensure it accurately reflects the scholarship and aligns with Virginia Tech’s Principles of Community. 
• Avoid using AI to generate exaggerated claims or language inconsistent with the actual findings.

• Use AI to check journal formatting, citation style, or metadata accuracy. 
• Use AI to create accessible summaries, including plain-language or multilingual abstracts, for broader audiences. 
• Verify that all AI-generated content is accurate and appropriate before submission or posting. 
• Ensure AI use complies with publisher and sponsor disclosure rules.

• Write all conclusions, interpretations, and responses to reviewers yourself; AI can help with phrasing or organization but cannot replace human reasoning.
• Maintain ownership of intellectual contributions and verify all AI-assisted outputs for accuracy. 
• Do not allow AI to write or reword reviewer responses in ways that alter meaning or tone. 
• Always disclose your role and AI’s role transparently in acknowledgments or cover letters if required.

• Follow journal or conference policies on AI use and disclosure. 
• Use only Virginia Tech–approved tools to handle documents containing unpublished or confidential material. 
• Verify that AI tools do not store or reuse text or metadata from manuscripts. 
• Do not upload drafts, reviews, or editorial correspondence to public or Virginia Tech-unapproved AI systems. 
• Do not use AI to conduct peer reviews, summarize, or critique peer-reviewed manuscripts.

• Review and approve all AI-assisted materials (e.g., cover letters, responses to reviewers, summaries) before submission.
• Retain full accountability for claims, interpretations, tone, and scholarly conclusions
• Do not delegate decisions related to authorship, acknowledgments, conflict-of-interest statements, or disclosure requirements to AI tools.

• Disclose AI assistance in cover letters, acknowledgments, or author contribution statements as required by journal or publishing entity. 
• Use AI to improve accessibility (e.g., translation, formatting) while ensuring human review. 
• Avoid language or translation tools that introduce bias or misrepresentation. 
• Be transparent with collaborators and co-authors about how AI was used during manuscript preparation, including translation if applicable.

• Use only secure, Virginia Tech–approved AI platforms for editing or formatting unpublished manuscripts.
• Avoid entering manuscript text, reviewer comments, or embargoed materials into public or unapproved AI tools.