SAFE Labs

BABLab_2026:

This is the part that matters most given we work with children’s and families’ data. UNSW has activated Microsoft Copilot with Commercial Data Protection for everyone with a zID — this is currently the only generative AI tool that’s appropriate for anything involving UNSW research data, because inputs aren’t used to train external models and are covered by UNSW’s data agreements. Consumer tools (a personal ChatGPT/Claude/Gemini account, browser extensions, random “AI note-taker” apps) are not appropriate for anything containing participant data, even de-identified data, unless you’ve specifically checked with Bridget and the lab manager first. When in doubt, treat it like you would any other cloud service you haven’t vetted — don’t put lab data into it.

IRIBHMComputationalLabs_2026:

We allow the use of AI-based services like ChatGPT or Grammarly to improve the quality of writing, however, fully AI generated text is not acceptable. It is also important to always verify whether the output is still correct, since large language models can give factually incorrect output.

CastoriniLin_2026:

If you ask a coding agent to do everything for you (and yes, a modern coding agent can easily speedrun the entire sequence), then it’s likely that you’ll miss much of the intended learning. Use the agent to help you, not to do everything for you. Remember, you are solely responsible for understanding what you’re doing and accountable for everything that you and your agent do.

EvendenLab_2026:

In the Evenden Lab, GenAI cannot be used as a ghost writer. It should not be used to generate code for statistical analyses or original written work or images. We do not upload our data sets to GenAI tools that share files. It can be used for troubleshooting and copy editing.

Disclosure/Transparency Policy:

• Keep chat logs with AI tools and notes detailing when generative AI tools were used
• Disclose any and all use of AI tools in research development (hypotheses generation) and writing with supervisor. Maintain transparency with supervisor about involvement of AI tools in analyses. See example disclosure statements here
• Ensure that AI use is within guidelines for funding agencies and journal requirements and its use is communicated clearly in published manuscripts and theses
• Individuals maintain ownership and responsibility for the interpretation, use and accuracy of generated outputs
• Ensure that AI generative tools are not saving your data and using that to train its models

GribbleLab_2026:

Using these tools to polish prose, catch typographical errors, or improve the clarity of sentences you have already written is generally acceptable. I sometimes use Claude to check this lab manual for spelling mistakes, typos, and grammatical errors.

However, in our lab it is not appropriate to use AI tools to perform the intellectual work that lies at the heart of our work as scientists. This includes synthesizing ideas from the literature, generating research questions or hypotheses, critically evaluating papers, drafting the content of a research paper, or grant proposal, or formulating arguments and interpretations. These kinds of tasks are precisely what you are here to learn to do, as scientists in training.

EMITLab_2026, translated from Chinese:

6. AI-assisted writing

AI tools can help to:

• structure arguments;
• improve language and expression;
• check consistency of terminology;
• turn existing facts into abstracts, figure legends or response outlines.

But all AI output must be checked by the author, who is responsible for:

• numbers, units and sample IDs;
• facts and citations;
• the strength of conclusions;
• whether any limitations have been left out;
• whether the AI has made anything up;
• the accuracy and completeness of the final submitted text.

Before using an external AI service, check that you have the right to upload the content; if unsure, ask the project lead.

EmbodiedComputationGroup_2026:

Record AI assistance accurately

Keep a note of the tools used, the tasks they assisted with, the data they could access and the verification you performed. Use that record when preparing the manuscript’s disclosure, following the journal’s instructions.

For example, if accurate:

We used Claude Code (Anthropic) to assist with analysis code. The authors reviewed the implementation and checked it using simulated data and independent calculations.

Only claim that no participant data entered the model if you have established that this is true. Do not copy a disclosure that describes checks you did not perform.

CzarkwianiLab_2026:

People are getting more and more creative with finding applications for AI tools, therefore, rather than specifying what people can use it for, I find it easier to list the unacceptable uses.

• plagiarizing writing or art
• directly copying AI output with no verification or validation (facts, citations, code, logic)
• substituting reading of full journal articles with AI summaries (in the context of journal clubs or reading vital literature for the research)
• inputting sensitive data to AI tools lacking appropriate data-protection policies

Reports, write-ups, and manuscripts really benefit from the human component in my opinion. Although I am not against the use of AI as a support tool it should not be used to fully replace you as an individual doing and writing about science or communicating with other humans.

INSIGHTLab_2026:

When such tools are used to support research activities, their role must be transparent, appropriate to the task, and aligned with open science and reproducibility standards. Clear distinctions between studying algorithms, applying algorithms, and using generative tools are essential to maintaining conceptual clarity and methodological integrity.

While there is no blanket ban on the use of AI tools in our lab, consider the following:

• Is it clear and transparent how the tool works? In order to justify using the tool for scientific work, we must know how it operates.
• What is the goal of the work you’re doing? In many cases, the goal is not to get the work done - the goal is learning how to do the work. For example, a PhD is expected to graduate with an academic level of competence in writing, coding, data analysis, and research skills. Outsourcing these tasks to an algorithm does not serve the purpose of learning those skills.
• Is the AI tool ethically trained? Just like we would not resort to sweat shop labor or copyright infringement to obtain research data, we should not use tools that use these methods to train an algorithm.
• Is the tool validated for the task you wish to assign it? Every tool that is used in scientific work must be validated; this includes measurement instruments, statistical methods, and of course, “AI tools”.

Generative AI may never be used for original writing, including papers, presentations, applications, and peer review. Generative AI may be used to suggest minor edits, like reducing word counts or simplifying language. The results should be thoroughly proof-read and edited.

BartomeusLab_2026:

We do not prioritize speed. Earning a PhD is learning how to do science. How to write code, how to write an introduction. This implies doing it yourself and not delegating your tasks.

• When the benefits of AI for a particular task are clear, we select the right tool (e.g. Google Scholar Labs for references) and use it with care. This often implies cautious prompts that are as specific as possible, and not broad task delegation. e.g., explain to me why this fails > fix this line of code > write me code that does X.
• Be upfront. Discuss early the expected genAI usage for a given project with all co-authors, and ensure everyone is comfortable with the plan. Disclose substantial AI use in all our outputs. If we use it, we explain it. This is transparency. Exceptions can be small grammatical edits by e.g., Grammarly, and small technical questions, but any paragraph, code, or image generated should be credited.
• We take responsibility for checking all AI outputs. It is our duty to fact-check AI work, from ensuring the code is functional, the references exist and say what the AI claims, or the suggested text corrections do not change the intended meaning. In a nutshell, we take full responsabiliyty of the tools we use.

SuperLab_2026:

You are discouraged from using large language models (LLMs) such as ChatGPT in your writing, including proofreading and refining language, summarizing research in literature reviews, and generating reference lists. LLMs “hallucinate” (fabricate) information and there are embarrassing precedents that we must not emulate in our own publications.

There is empirical evidence that using generative AI reduces your frontal cortex function.

Reshanne will not be a co-author on any work that uses generative AI in any way.

AeryJonesLab_2026:

• AI tools may be used for some tasks in our lab with caution and all outputs must be verified.
• Code: AI can be used to assist in writing code (e.g. Github Copilot) if the outputs are thoroughly tested, as you would for any manually generated code. AI cannot be used to write large amounts of code independently (”vibe coding”), as it will likely contain critical errors. All code must be well formatted and documented.
• Transcription: AI may be used for live transcription of meetings if necessary for accessibility and may be used to generate meeting summaries by tools that immediately delete recordings (e.g. Fireflies). Everyone in the meeting must consent first.
• Literature Review: AI can also be used to identify relevant papers for you to read. You must read these papers yourself rather than taking the tool’s summary as fact; these summaries usually have serious flaws in them.
• Writing: AI may be used to suggest minor edits, like reducing word counts or simplifying language. AI may never be used for original writing, including papers, presentations, applications, and reviews, as these should be in your own words.

AllisonAdolescentResearchLab_2026:

In general, I prefer that you develop initial versions of all study documents and obtain input from the study team, and then am comfortable with you using generative AI to refine protocol documents, interview guides, participant or eligibility surveys, recruitment materials, clean deidentified transcripts, or consent language for the IRB. I fully expect that you do your due diligence and take the time to carefully review any content generated from AI, and then edit it to accurately reflect the work we do. You should never put any identifiable information into generative AI. When it comes to scholarly products and dissemination of our research (e.g., manuscripts, presentations), I am more cautious about if and how AI can be used due to difficulties maintaining research integrity and recognizing infringements on others’ intellectual property. We do not want to inadvertently propagate misinformation, or worse – engage in any research misconduct. Do not use AI in the preparation of scholarly products for dissemination (e.g., editing text for grammatical correctness or clarity) without discussing it with me first. Be aware that publications, medical organizations, and conference organizers have different norms and policies on if/how AI can be used to prepare scholarly contributions. Unacknowledged or inappropriate use of AI in manuscripts, grant applications, or presentations is considered plagiarism/ misconduct.

AneGroup_2026:

Generative AI is incredibly useful. Overarching guidelines:

1. use it to be more productive
2. question everything, check accuracy, take full 100% responsibility for the outcome
3. don’t use it to skip learning
4. never input confidential information, such as data or unpublished manuscripts

For #2 specifically:

• check any code produced by generative AI: accuracy is paramount
• review all writing produced by genAI: you will be the author in the end
• as everything, document: before you start using AI, document the tool version, document how & why you use it, save your prompts. You will need this information.
• disclosure: write an AI statement to disclose your use of generative AI, to be included in any resulting publication.

BenjaminChungLab_2026:

Lab members must disclose GenAI use in research at four points:

1. Within the lab, to Jade using the Benjamin-Chung Lab AI Use in Research Documentation Form
2. To collaborators, whenever their data or materials are involved
3. To journals, at time of submission (see below)
4. To funders, when their policies require it.

DevPsyLab_2026:

All AI use must be disclosed if used for research/external products (i.e., anything that will be used or viewed outside the lab). Research products (i.e., posters, manuscripts) should include a statement describing the use of AI using the categories below:

• Topic Selection: using text generators like ChatGPT to help brainstorm potential research topics to study
• Brainstorming: using mind-mapping tools to help connect disparate ideas or suggest related concepts
• Literature Search: using ChatGPT plug-ins like Scholar AI or websites like Perplexity AI to assist you in conducting scholarly reviews of the literature (always ensure sources are valid and read any articles you cite; these tools sometimes create false citations)
• Outlining: using a text generator to help organize thoughts, main arguments, and supporting evidence into a coherent outline
• Feedback on writing: using a text generator to analyze the draft for grammar checking, clarity, coherence, and consistency, or to help polish or refine language, syntax, tone, and overall style
• Programming: using AI tools to help write code or syntax
• Analysis: using AI tools to help with data analysis (besides mere programming)

Here is an example attestation for a research product that used AI:

“I have reviewed the lab’s AI policy and attest that I fully report (below) the ways I used AI: Literature Search, Programming, and Analysis. I attest that I am responsible for the content of this project.”

Here is an example attestation that did not use AI:

“I have reviewed the lab’s AI policy and attest that I did not use AI for creating this project. I attest that I am responsible for the content of this project.”

EcometeorologyLab_2026:

The advances in generative AI transformer-based large-language models (LLM), agentic AI, and related tools such as ChatGPT, Claude, Gemini, CoPilot, and so on, plus derivative tools like Grammarly, Codex, or translation engines, are evolving rapidly and require paying careful attention to their limitations, ethics, and resources required or cost. In their current form, these tools are great for idea generation, automating routine workflows, debugging or generating code, developing figures or interactive web apps, overcoming writer’s block or other academic struggles, improving your English (or other language), interpreting an email, or helping summarize or improve text. However, you need to understand the science, code, and text generated, potential security risks, and be able to explain the theory and understand where AI use is allowed or prohibited. The risks of hallucinations (made up answers), false citations, overly confident sounding yet incomplete answers, and incorrect or incomplete math or code means they are require thoughtful use for scholarly output including approaches to verify output. If you are unsure of whether an AI tool is appropriate in a specific context, ask me first. You may be violating intellectual property agreements if you feed into it academic work (e.g., a manuscript you are reviewing), and many journals and funding agencies prohibit its use in that way or require disclosure when used for parts of manuscripts. The university licenses AI tools that may be more appropriate, secure, and cheaper than commercial ones, so check those out first.

GureckisLab_2026:

The following checks are mandatory before any manuscript submission. Citations. Verify that every DOI resolves, every URL is live, and a human has read at least the abstract of every paper cited to confirm it is being accurately characterized. Statistical code. Any analysis generated or substantially assisted by an AI tool must be independently verified on simulated data with known ground-truth properties before application to real data. The verification script must be preserved and submitted with the code repository. Model implementations. Cognitive model code must include unit tests covering known closed-form cases, boundary conditions, and, when possible, an end-to-end test reproducing a result from a prior published paper. Tests must pass before any data is published with the model. AI-generated prose. Every factual claim in AI-assisted text must be traceable to a primary source. “The AI wrote it” is not a source.

GWLab_2026:

Use of generative AI/large language models: LLMs are a rapidly evolving tool that can be helpful but can lead to serious environmental, ethical and legal violations. Most importantly, LLM use can disrupt the learning process. As we are all learning, shortcutting this process robs us of opportunities for growth. We are scholars and the purpose of scholarship is to help more people better understand the world not to generate text. As such, our default position is that AI use for scholarship (in the lab) is not acceptable, with exceptions only as approved by Sarah.

• Never upload data into an LLM/online AI tool. The data is owned by the lab and doing so violates our intellectual property rights. LLMs are not good at data analysis and will often find patterns in data that aren’t there.
• Never upload an unpublished manuscript or grant proposal into an LLM/online AI tool.
• AI for any sort of scientific writing is not acceptable. The act of writing is the act of thinking; using AI to write is bypassing thinking.
• AI tools such as grammarly to help with wording/grammar is only acceptable in certain situations and after conversations with and approval from Sarah.
• AI for help with coding is acceptable in certain situations after approval from sarah: the output of the LLM must always be verified and tested, and use of an LLM must be acknowledged in the code comments. You must do the work to be able to explain and stand behind any code you put out into the world.

HewittLab_2026:

Writing and verbal communication are core research skills we need to practice. All paper first drafts must be written without AI. Lab communication must follow the first-follow-up rule: lab members should know enough details about the code, experiments, and presentations they give to answer a reasonable first follow-up question from a labmate without checking with their AI assistant.

KirbyLab_2026:

1. First, apply the “Other Person” rule. Ask yourself: Would it be ok to pay another person to do what you are trying to use the LLM or AI program for? If so, it is possibly ok. If not, then it is probably not OK. Some examples: • Would it be ok to pay another person to re-write your thesis in better prose? Nope. That’s obviously plagiarism, even if the other person is just paraphrasing your writing. So, it’s not ok to ask any LLM to do that for you. • Would it be ok to pay another person to design your experiments for you or to come up with your next experiment? Nope. Everyone here is in training and part of what you are training to do is design experiments. Paying another person to design or conceptualize your experiments would be side stepping your actual training (and

probably violating authorship policies if the other person is not credited). Plus, this one would be hard to execute without feeding our unpublished data into the LLM. • Would it be ok to use AI or an LLM to summarize primary literature for you? Hard no. Learning to read and interpret primary literature is central to training in research. Paying someone else to do that would be unacceptable, and so would using an LLM to do it.

LoaizaCAMLab_2026:

The use of artificial intelligence (AI) and the internet more broadly (e.g., web searches via Google) to help you find solutions to problems when it comes to things like coding experiments/analyses, preparing study materials, or thinking of how to rework sentences in your writing is absolutely fine when it comes to the lab. However, you should never let the resource do the work for you. In other words, avoid mindlessly copying/pasting the results of your searches/prompts to what you are working on. Instead, you should consider whether the output is actually relevant/helpful, revise the information according to your needs, and double-check other sources to be sure that the information is accurate. In short, AI and the internet are helpful tools, but they are no replacement for our big, beautiful brains. If you are ever in doubt, please ask Vanessa – open and non-judgmental discussions can be important to determining what makes the most sense for what you are doing.

NicholGroup_2026:

APS, NSF, and NIH all have guidelines about appropriate use of AI. The primary concerns are the following.  An AI cannot be held responsible for any errors. As such, it cannot be credited with authorship. This means that any errors introduced through the use of AI, such as plagiarism or incorrect calculations or citations, are the responsibility of the human authors.  Any information you upload to an AI is effectively in the public domain. At a minimum, it means you cannot upload confidential documents like papers or proposals that you are reviewing, for example. However, it also means that the draft of the paper you just uploaded for editing is no longer confidential. My current guidance is the following:  Substantial parts of papers should not be uploaded to an AI before publication.  If you use an AI for any part of the writing process, please let me know.

PeixotoLab_2026:

When it comes to scientific writing, below are not acceptable and acceptable uses of generative AI tools in the lab:

Not-acceptable uses:

• First draft of a scientific manuscript
• First draft of a grant proposal
• First draft of an abstract or summary.
• Literature search, without independent verification. It is OK to use AI to enhance your own search but remember that AI tools are prone to hallucinating references. You should always read the references AI suggests (or any references you cite).
• Anything that involves uploading identifying information. WSU Executive Policy 8 prohibits the inclusion of legally protected or regulated data (e.g., proprietary, personally identifiable information, HIPAA, FERPA) in queries provided to generative AI platforms.

Acceptable uses:

• Brainstorming/outlining
• Summarization (e.g., shortening length of an abstract)
• Enhanced writing and editing (e.g., grammar checks)
• Filling of pre-formatted forms or tables
• Critical feedback of human-written first draft.
• Code debugging.

I require disclosure of the use of AI in manuscripts and grant proposals in writing by email to me before submission, with an explanation of what tool was used and what for. Several journals now require disclosure at submission as well. If this disclosure is found not to be honest or violate any of the provisions set above (e.g. if hallucinated references are found in the manuscript text), it will be considered scientific misconduct and may lead to removal from the manuscript authorship list or withdrawal of the grant proposal.

PoldrackLab_2026:

AI tools may be used for any part of the research process except for the drafting of narrative text for manuscripts or grant proposals. Lab members are expected to manually draft all of the text that they put into a paper or proposal; no copying and pasting of text directly from an LLM is allowed without explicit discussion and agreement.

When using AI tools in the research process:

• AI tools may be used to create schematic figures but the accuracy of the figure must be double-checked and confirmed.
• AI tools must not be used to directly generate data figures; instead, the AI tool should be used to generate code that will create the figure in a reproducible and auditable way.
• AI tools should not be used to generate reference lists.

RaredonLab_2026:

Make your own slides. 1:1 slides must be made by you, not generated by AI. These slides are a record of your thinking. The PI wants to see how you understand your work, what you notice, what you choose to highlight, and where you are uncertain. An AI-generated summary cannot show any of that. It obscures the very thing the PI needs to see in order to help you.

RobertsLab_2026:

The goal is not to discourage AI use, but to make its role transparent. Authors remain responsible for the accuracy, originality, interpretation, reproducibility, and ethical use of all material in the final work product.

Each work product should include one AI Use Indicator level, selected by the author based on the highest level of AI involvement. For example, if AI was used for both grammar editing and code troubleshooting, the appropriate level would likely be Level 2 rather than Level 1.

A recommended placement is near the end of a notebook post, README, manuscript acknowledgments section, presentation title slide, or GitHub commit message.

Level	Label	Description	Short disclosure
Level 0	No AI use	No generative AI tools were used in creating the work product.	No generative AI tools were used in creating this work product.
Level 1	AI-assisted editing	AI was used only for grammar, clarity, formatting, summarizing, or minor wording improvements. The ideas, analysis, code, and interpretation are human-generated.	Generative AI was used only for grammar, clarity, formatting, or minor wording improvements. The ideas, analysis, code, and interpretation are the author’s own.
Level 2	AI-assisted drafting or coding	AI helped draft sections, suggest code, outline content, troubleshoot errors, or reorganize material, but the author substantially revised, verified, and approved the final product.	Generative AI was used to help draft text, outline content, suggest code, or troubleshoot errors. All outputs were reviewed, revised, and verified by the author.
Level 3	AI-supported analysis or interpretation	AI contributed to analytical decisions, interpretation, figure design, literature framing, or synthesis of results. Human review and verification are required and documented where appropriate.	Generative AI was used to help evaluate analytical options, interpret results, design figures, or synthesize information. Final decisions, validation, and conclusions are the author’s responsibility.
Level 4	Substantial AI contribution	AI produced a major first draft, substantial code, synthetic interpretation, or large portions of the work product. The final version required human validation, revision, and responsibility.	Generative AI produced a major first draft, substantial code, synthetic interpretation, or large portions of the work product. The final version was reviewed, revised, validated, and approved by the author.

RossIbarraLab_2026:

The expectations differ by experience level: what is appropriate for an expert may not be appropriate for a beginner, because part of training is learning to do the work yourself first. Your relevant experience level will likely differ for different tasks, so please use your best judgment and ask for clarification if you are unsure. The goal is to ensure that you are learning the skills you need to be a successful researcher, while also using AI tools responsibly.

General Best Practices

These principles apply to everyone in the lab, regardless of experience level:

• Always check AI output before relying on it.
• Ask for sources and verify them.
• Accountability — independently validate what the AI produces.
• Ask questions about assumptions the AI (or you) may be making.
• Ask before sharing others’ writing and code with AI tools.
• Feedback from a human will almost always be more helpful than AI.
• When in doubt, ask someone!
• Should you learn it? If knowing this is important to your development, do it yourself. If not, you can use AI.
• Generally: Could you do it without AI? If not, do not use it until you can.
• Ask for point-by-point review of text, rather than have AI rewrite anything.
• Ask to see code rather than blindly accepting it.
• Reading, interpreting facts, and synthesizing content are core scholarly skills you are expected to develop and own.
• AI should never outline a draft or rewrite your writing — the thinking embedded in drafting and writing is yours to do.

WeisenseeLab_2026:

Before you send me any written drafts & documents I expect you to carefully proofread the text,  captions, and figure legends. If your native language it not English and you struggle with grammar and  the “flow” of writing in English, I highly encourage you to work with the Engineering communications &  writing  center  to  edit  your  documents  before  you  send  them  to  me  for  review  or  to  use  artificial  intelligence (AI) for language editing (be careful about AI hallucinations, though!). I expect that we will  have to go through numerous iterations between you and me before manuscripts are ready to submit,  but  these  should  focus  on  the  scientific  content  and  storytelling,  not  English.

ZETAResearchGroup_2026:

For us to have a successful advisor-advisee or mentor-mentee relationship, we need to be able to trust each other. As part of that, I expect for you to disclose any use of a LLM to me. Failure to do so, jeopardizes that collaborative relationship and may mean we are not able to trust each other. For disclosing LLM use, please complete the following table (or a modified version we agree upon). This table must be completed prior to using an LLM.

Task	Rationale	Description of Use
Outlining project	…	…
Writing proposal or paper	…	…
Statistical analysis plan	…	…
Code writing	…	…
Data analysis	…	…
Code or paper review	…	…
Literature review	…	…
Other	…	…

I also recognize that the mentor-mentee relationship is a two-way street. So, I also make the following promises in terms of conduct and disclosure to you:

• I will review your work that you send to me. I will not use an LLM or other related technology for reviewing your work.
• I will respond to your emails or questions myself. I will not use an LLM to do so.
• I will draft requested recommendations letters myself and will not use an LLM.
• If we are working together on a project, I will disclose my use of an LLM in accordance with the disclosure table.