Skip to main content

← AI ethics and policy

FAQ on AI, Research Integrity and Data Safety

by Sergueï Spetschinsky

The guidelines behind these answers, and how I approach them, are in my note AI, Research Integrity and Data Safety.

Last updated: 7 November 2026.

Written first for doctoral researchers; most answers hold for any researcher. This is general information. I am not a lawyer, and your own institution, journal and funder come first. I teach researchers to work with AI, for a fee, and I built some of the tools mentioned here (see the foot of the page).

When I say most, I mean the published rules of 39 of 50 named European universities, and of the funders, publishers and learned societies I analysed, as they stood on 4 November 2026.1 American universities appear as examples and are outside those counts.2

Start here.

Where the rules stand

What European university documents say, act by act, as I analysed them. These are mostly rules for students and researchers in general, because few speak to doctoral work directly.1

What you do with AI Commonest position Counted
Spelling and grammar No declaration needed 10 of 19 documents; 6 want it declared
Rephrasing and style Declare it 11 of 20
Summarising literature Declare it 11 of 16
Brainstorming and critique Declare it 7 of 10
Outline and structure Declare it 7 of 11
Drafting text Divided 16 of 28 say declare it, 10 forbid it
Writing code Declare it 8 of 12
Personal data in a tool Forbidden, or only with permission 13 of 15
Reviewing someone else's work Forbidden 8 of 9

The rest leave the matter to your lecturer, supervisor or faculty, or do not speak to it.

Is it allowed?

Am I allowed to use AI in my thesis?

For support tasks, such as summarising literature, brainstorming, proposing an outline, improving your style or writing code, usually yes, if you declare it. Having AI write text for you, even as a draft, is where the rules divide (see May AI draft my text or propose my outline?). None of the 39 European universities whose texts I analysed has a general university-wide ban, and most of their texts allow AI if declared or leave the decision to your lecturer or supervisor. Only 19 reach doctoral work at all, often in a faculty's rule or on a form you sign.1 Heidelberg, for example, allows AI in a dissertation if "explicitly disclosed, specifying when, where, why, and to what extent".3

Some universities and faculties are much stricter. Harvard's Graduate School of Arts and Sciences says: "Unless a department or advisor explicitly authorizes in writing the use of generative AI tools, it is assumed to be prohibited for a student's academic work."4 MIT's electrical engineering and computer science department expects the author to write the original drafts, allows AI to improve the grammar of a final draft, and does not allow it "to compose text starting from bulleted text".5 Cornell's PhD programme in Communication says: "Students may not use GenAI platforms such as ChatGPT to draft or write their A exam papers or dissertation."6 Oxford's Classics Faculty bars AI for producing the text or shaping the argument of assessed work, theses included, with a permission route for AI as a research method. Oxford's research policy for the whole university allows use if declared, and sends assessed work to a separate policy.78 Innsbruck's Faculty of Catholic Theology forbids AI-generated text in written work.9

The doctoral declaration of Basel's Faculty of Law and Innsbruck's theology rules both name withdrawal of the title or degree.109

Before you start, find your faculty's rule and the declaration you will sign. If you find neither, ask your supervisor, in writing, and keep the answer.

Who decides what I may do with AI?

Usually your faculty, your lecturer or your supervisor, within a frame your university sets. For teaching and assessed work, Zurich says its faculties write the concrete rules.11 At Cornell, graduate students follow the norms of their field, which "vary and continue to evolve by discipline and, for research degree students, by special committee"; at Berkeley, the program or the instructor sets the terms.1213 Few of the European texts name the supervisor: 21 of the 142 university texts I analysed.1 I would start there anyway, because that is the person who will read your thesis.

My university, my funder and my journal say different things. Which one wins?

I found no text that ranks them. I would follow the one that governs the work in front of you: your journal for a paper, your funder for a proposal, your faculty for the thesis. Where two apply to one act, I would follow the stricter. The EU's guidelines, which do not bind, say of themselves: "Where specific legal or policy requirements apply, these take precedence."14

My university has no rules on AI. What do I do?

Read what it does have. In most of the cases I analysed, that means rules for teaching or research staff that do not name doctoral candidates. Then work to what the rules elsewhere share:

  • You answer for the result.1516
  • AI is not an author.1718
  • You declare any use that shaped the work.1920
  • Confidential and personal data stay out of open tools.2122

Ask your supervisor what they expect, in writing, and keep the answer. I would keep a log from the first day.

Can using AI in my research break the law?

Yes, depending on what you do with it. Here are the cases I would watch.

  • Putting data about people into a tool. Data-protection law applies; the Swiss data-protection commissioner calls it directly applicable to AI.23 Under the GDPR, the EU's data-protection law, doing it with no legal basis, or with no contract with the provider where one is required, can breach the law (see May I put interview or participant data into an AI tool?).24
  • Uploading other people's texts. Copyright law in your country, publishers' licences and confidentiality can forbid it (see May I upload a published article or a library PDF to an AI tool?).
  • Recording people to get a transcript. In Germany and Switzerland, recording someone's non-public words without permission can be an offence (see Are Zoom, Meet, cloud transcribers and dictation tools a data transfer?).2526
  • What you sign and submit. Exam rules and declarations carry legal weight. German administrative courts have decided against students over AI: Kassel on two exam papers, one a bachelor thesis, reported in March 2026 as not yet final, and Munich in an interim order on an admission essay. None concerns doctoral work.2728 For Austrian institutions, the higher-education quality act lists abusive use of AI among the unauthorised aids that count as scientific misconduct.29

The EU's AI Act excludes systems built solely for research, and its recital says any other system used in research stays subject to it; it does not name chatbots.30 For your own case, ask your university's data-protection office or legal service.

Reading and finding literature

May AI summarise the papers for my literature review?

Usually yes, if you declare it and check it. Of 16 university documents that speak to it, 11 allow it if declared and none forbids it.1 A summary tells you what to read next. I would read whatever you are going to cite.

May I upload a published article or a library PDF to an AI tool?

This is not settled, and few places have a rule on it. Yale's library says: "Copyright law in generative AI contexts is far from settled."31 Those that do speak differ. Helsinki goes furthest: "It is against copyright law to use copyrighted material, such as copyrighted publications, as an input to GenAI tools."32 Oxford says uploading "may constitute a breach of copyright" unless an exception covers it.8 Columbia warns that "some publishers are asserting that using their content with AI tools is not allowed".33 Fribourg's library, for the journals it covers, allows licensed content in closed tools that do not train on the input, or in a self-hosted or access-restricted environment.34

Ask your university library what its licences allow. Even open-access papers carry a licence, so check it: at Fribourg, content under a CC BY licence may go into any tool, and content under an NC-ND licence is restricted like licensed content.

How do I catch a reference AI invented?

Look it up yourself, in a library database or at the publisher. Do not ask a model: in one test of 2025 models, 13 of them judging 100 references, half invalid, averaged 38% accuracy.35 Heidelberg tells its doctoral candidates that AI-suggested references must be individually verified.3

I built a free skill for Claude, hallucination-check, that looks each reference up in the scholarly databases and marks those it could not find or could not check: download it here. It is far from reliable, so I would use it only to see which references to check by hand first, never to decide whether a reference is real.

How do I cite an AI tool?

Declare it. Whether you also list it as a source differs. Princeton says AI "is not a source, since its output is not produced by a person", and Heidelberg says AI systems are not valid sources or authors and "must not be cited as such".363 Princeton itself notes that the Modern Language Association expects much of the usual bibliographic information.37 Check your style guide and your faculty, and ask your supervisor.

May I use AI to translate someone else's text, or to read sources I cannot fully check?

It varies, and only six of the university documents I analysed speak to it, too few to show a common position.1 Oxford counts translation as substantive use, to be declared.8

In a language you cannot check, I would treat the output as a lead and have whatever you quote read by someone who knows the language. Ask your supervisor how they want this declared.

Thinking with AI

Is it still my work if AI only helped me think?

You answer for it either way, and most rules want it declared. Of 10 university documents on brainstorming or critique, 7 allow it if declared and none forbids it.1 If it changed your argument, I would say so.

How do I tell when AI is simply telling me what I want to hear?

Test it, because these tools can lean towards pleasing you. I found no rule on this, so this is my own advice:

  • Ask for the strongest objection rather than for an opinion.
  • Give it a version of your argument with a flaw you planted, and see whether it finds it.
  • Keep a human reader you trust.

Will I lose research skills if I lean on it?

It depends on what you hand over. I found no rule on this. Germany's science council recommends keeping spaces in degree curricula where demanding independent thinking can be practised without AI, and its press release does not address doctoral research.38 I would decide which skills your doctorate is meant to build, and practise those by hand.

Writing with AI

May AI draft my text or propose my outline?

It depends on your rule. Of 28 university documents on drafting, 16 allow it if declared and 10 forbid it; the bans sit mostly in a faculty, a department or coursework.1 Outlines: 7 of 11 allow it if declared. Heidelberg lists text generation among legitimate uses, if disclosed. Oxford's Classics Faculty, unusually, prohibits even "proposing introductions, argumentative structure, or conclusions" in assessed work.37 Stanford's default rule for courses, which an instructor may change, says students "may not use AI tools to create deliverables".39 Columbia allows it if disclosed, and says its members "must not hold out any output generated by Generative AI tools as their own".33

I would keep the argument and the core of the writing yours, and talk to your supervisor before AI writes anything for your thesis.

Is lightly reworded AI text still plagiarism?

I would treat it as AI text, and declare it. Few texts speak to light rewording as such. Princeton's history department tells its graduate students: "You should never copy and paste text produced by generative AI, even with modifications, and submit it as your own work".40 Elsewhere the question is whether you said so: an administrative court in Kassel, on AI-generated passages in student work, reasons that making a scientific text with AI without saying so already breaches good scientific practice (reported in March 2026 as not yet final).27

How far may AI edit my text?

Further than spelling, usually, if you declare it. Of 20 university documents on rephrasing and style, 11 allow it if declared and 6 need no declaration.1 The rules differ widely. KU Leuven needs no mention when a model is used only as a language tool. Innsbruck's Faculty of Psychology wants even simple correction declared, and Cornell's PhD programme in Communication wants language and grammar help disclosed.41426 Where your thesis rules are silent, ask your supervisor how far to go, and keep the answer.

Do I have to declare help with grammar, spelling or translation?

For spelling and grammar, often not. For translating your own text, it is split. Of 19 university documents on spelling and grammar, 10 need no declaration and 6 want one; of 10 on translation, 5 want it declared and 4 do not.1

The EU's guidelines say polishing generally needs no disclosure, and add: "In case of doubt, disclosure is advised."14 Go by what the tool did to your text. A declaration takes one sentence.

I have dyslexia or another support need. May I use AI for it, and what do I declare?

Ask your university's disability office first, because I found very few rules on this, and all of them British. Dundee says: "Students using AI for reasonable adjustments recommended by Disability Services are NOT required to indicate use".43 The Open University allows it as a reasonable adjustment, if it fits the assessment's learning outcomes.44

I found no German, Swiss or Austrian text that links AI to an adjustment for disadvantage (Nachteilsausgleich), so I would ask for it to be named in yours. I would also ask your office how to word a declaration that does not reveal your diagnosis.

May I dictate to AI and let it turn my speech into text?

I found no rule on dictating your own words, and I write this way myself. If the tool only transcribes, I would count the words as yours; if it rewrites, treat it as editing. If you dictate anything about other people, see What should never go into an AI tool?

If I teach an assistant my writing style, am I hiding something?

Not by itself, in my view, if you declare the use. I found no rule that mentions it. The European code of research integrity lists "Hiding the use of AI or automated tools in the creation of content or drafting of publications" as unacceptable.45 So the declaration decides on hiding. Whether AI may draft your text is a separate question (see May AI draft my text or propose my outline?).

Do humanising tools work?

These tools rewrite AI text so that detectors do not flag it. The evidence is thin and dates quickly. A working paper of 2025 finds that the best commercial detector holds up against them.46 It does not test human readers. I would not use one: hiding AI use is what the European code lists as unacceptable, whether or not a detector notices.45

Building your own assistant

May I build an assistant from my own drafts and notes?

I found no rule for or against it. This is much of what I teach, so weigh my view with that in mind. Helsinki wants the tool's name, version, date of use and how it was used, and says not to use unpublished results as input.32 I would declare both the assistant and what it produced.

On personal plans, uploads may train the model unless you switch that off. Tell your supervisor what you have built. I would show them.

May AI read my chapter the way my supervisor would?

I found no rule that forbids it, and most texts treat critique as use to declare: 7 of 10 university documents.1 It prepares you for a reader, and it does not replace one: Leiden's law school says even supervisors may not rely solely on AI for feedback on doctoral work.47 So I would tell your supervisor.

I found no rule that speaks to this directly, and I would be careful. A mailbox and a shared drive hold other people's data, so data-protection law and your university's input rules reach whatever the tool reads.48 Linking a mailbox or shared drive can send other people's messages to the provider, so I would not do it on a university account without written permission from your IT service. I would start with a folder holding only your own material.

Data, code and figures

May I use AI to code or analyse qualitative data?

Few texts speak to it, and none names qualitative coding. Of 8 university documents on data analysis, 4 allow it if declared and 1 forbids it.1 For interviews, see What should never go into an AI tool?

Ask your supervisor, in writing, whether they accept it for your thesis. I would read the passages behind every code myself.

May AI write the code for my analysis?

Usually yes, if you test it and declare it. Of 12 university documents on writing code, 8 allow it if declared and none forbids it.1 Taylor & Francis, a large academic publisher, warns that code generated without rigorous revision "may be subject to editorial investigation".49 MIT's research guidance counts analysis and code development among the uses to disclose.20 I would keep the prompts and outputs, and run the code first on data whose result you already know.

Can I use AI for figures or images?

It depends on where the image is going, and publishers divide. Of 7 publisher documents, 3 forbid AI images and 4 allow them if declared. Of 10 university documents, 6 allow them if declared and 1 forbids them.1

Ask your journal's editor before you submit, or your supervisor for a thesis. I would never let AI alter an image that serves as evidence, such as a photo, a scan or a plot of real data. A chart your own code draws from your data is another matter, in my view.

Does AI carry biases into my analysis, and must I say so?

Assume it can, and I would say how you checked. Yale warns that AI responses "can be biased, inaccurate, inappropriate".16 The rules I found speak of responsibility and checking rather than of a duty to report bias.15 In your methods, I would name the tool, what you asked of it and how you tested the result.

Should I care about the environmental cost?

Yes, I would, and in practice it is a matter of choosing the tool. Oxford's research policy says: "Be mindful of the environmental impact of using GenAI tools".8 I would use the smallest tool that does the job.

Declaring and keeping a record

How much AI help do I have to declare?

It depends on your rule, and the lines differ widely. Innsbruck's Faculty of Psychology wants even simple text correction declared, while the Humboldt Foundation, which funds international research fellowships, asks its applicants for no disclosure for the time being.4250 Between them, declaring is the commonest position, in 35 of 67 European university documents on AI use in general.1 Penn's guidance says: "Disclose when a work product was created wholly or partially using an AI tool".51 The EU's guidelines, which do not bind, count a literature review or interpreting data analysis as substantial use, to be disclosed.14

Ask your supervisor, in writing, where yours sits. If in doubt, I would declare.

What do I write in a declaration, and where does it go?

It depends on where the text is going, and no single place or form is the norm.

  • A thesis. Six doctoral texts name the signed declaration. Others point elsewhere: Zurich's veterinary faculty wants a log as an addendum, and KU Leuven a description in the dissertation.15241 MIT's task force recommends a statement in every thesis, and Cornell's PhD programme in Communication wants a "reflexive commentary" in an appendix, with the prompts.186
  • A paper. Your journal decides. Elsevier, a large academic publisher, wants a separate titled declaration above the references.53 For research in general, Columbia points to "describing methods, acknowledgements, or elsewhere, as appropriate".33

For the content, Helsinki asks for "the name of the tool and the version used, the date of use, how the tool was used".32 Heidelberg asks for "when, where, why, and to what extent".3 Princeton offers a template that names the tool and the purpose, with page numbers for generated text.37

Use your own form where there is one. I would add how you checked the result.

Is one general note enough, or must I name each passage?

It depends on your rule, and both kinds exist. Basel's Faculty of Law has doctoral candidates declare that every passage written wholly or partly by AI is marked, with the program named.10 MIT's task force recommends one statement in the thesis about how AI was used, and Princeton's template names the tool and the purpose, with page numbers for generated text.1837

Where your rule is silent, ask your supervisor. If AI rephrased the whole text, I would say so once and describe how. If it wrote passages, I would mark them.

AI touched every paragraph of my text. What do I declare?

All of it, and first find out whether that much is allowed. I found no European doctoral rule written for this case. Harvard's Graduate School of Arts and Sciences does speak to it: where use "reaches a level that obscures the student's own intellectual contribution", it "may be grounds for failure and/or may lead to disciplinary action by the Administrative Board, regardless of disclosure".4 For coursework, Cologne's Philosophical Faculty treats work mostly generated with AI as an attempt to deceive, an unusually strict rule.54

Talk to your supervisor before you submit, and keep their answer in writing. A declaration does not by itself make the amount acceptable. I would describe the whole workflow.

Will declaring my AI use count against me?

Not by any rule I found, though I cannot promise how a given reader will react. The consequences I found fall on not declaring.27 A declaration does not make acceptable an amount of use your rule does not allow (see AI touched every paragraph of my text. What do I declare?). Researchers do fear it: in an Oxford University Press survey of 2,606 respondents, as Times Higher Education reports it, 44% worried that disclosing would hurt how others view their work.55 That measures fear.

The rules do not settle every case yet. What is clear: answer for what you sign, and do nothing in secret, because that is where people get into real trouble. Others may declare nothing. Your declaration speaks for your own work, and you are the one who signs it.

Can AI be listed as an author, or thanked?

Not as an author. Every text I found that speaks to it says no. The University of Bern's binding research guidelines say AI systems cannot be recognised as authors or co-authors, and MIT's task force writes: "AI should never be listed as a co-author."5618

I found no rule that asks you to thank an AI or forbids it, so I would declare the use and leave the thanks out.

Must I acknowledge AI more specifically than help from a proofreader or a colleague?

Often yes, though I found no rule that says so in those words. Dundee calls acknowledging a proofreader "a normal courtesy", and wants "a clear statement explaining how generative AI was used".5743 Princeton gives the reason: AI "is not a source, since its output is not produced by a person", so you disclose it instead of acknowledging it.36

I would thank the people in your acknowledgements and declare the tools in your declaration.

Do I have to keep a log of my prompts?

It varies, and most texts do not ask. Of the 205 documents I analysed, 8 ask for the prompts or the exchange itself, 10 recommend saving them and 153 mention no record.1

Zurich's veterinary faculty is unusual: it wants every prompt and answer logged for doctoral and master theses.52 Bern's university-wide research guidelines require transparency about the prompts used, without saying whether that means a kept log.56 Cornell's research task force says: "DO keep documentation and provide disclosure of GenAI use in all aspects of a research process".58

Check your faculty's rule. Where it is silent, keeping a log is your choice. I strongly advise it. I built a free tool for this, ai-log, a ready-made skill for Claude: download it here.

What does a good log of AI use look like?

I found no text that sets a template. Yale's library suggests recording the model and version, the company, the prompt, the output, the date and how you used the output.59 Here is what I would record for each use:

  • Date.
  • Tool and version.
  • Purpose, and for which part of the work.
  • What you gave it. The prompt itself, with anything about other people left out.
  • What came back. The answer, or where it is saved.
  • How you checked it.
  • What you did with it. Kept, changed or discarded.

If your faculty sets its own format, use that. A log is your raw record. It does not replace the declaration, which says what you used and how.

Your supervisor and your examiners

Must I tell my supervisor that I use AI?

Where your rules say so, yes, and I would tell them in any case. I found no rule that requires it in every situation. The stricter ones want it first. Yale's Graduate School makes it the student's duty to consult the relevant faculty member. Cornell's PhD programme in Communication has students discuss their AI use with their committee at each stage of their A and B exams. Harvard's PhD programme in Biological and Biomedical Sciences has student and adviser co-sign an attestation, renewed every year. Innsbruck's Faculty of Education says use in theses must be agreed with the course leader or supervisor.6066162

If your supervisor has no view, follow your faculty's rule, declare what you used, and keep a note of what you told them.

Will my supervisor think less of me?

It depends on the person, and I found no study that measures it. In one UK survey of 2025, of 308 social-science PhD students who chose to answer, 20% had ever openly talked about AI with their supervisory team.63 I would say what you use it for and what you check yourself, then ask how they want it declared. A supervisor who does not use AI may find this hard to picture, so I would show one concrete example.

What if my supervisor says no to AI?

Then I would treat the no as holding for that work, and using AI on it in secret is where the trouble lies. Graz's framework for study and teaching calls unmarked AI use, or use against the requirements set for the work, a breach of good scientific practice, and says theses follow the supervisor's instructions.64

I found no rule that says what follows a refusal. I would ask which tasks the no covers, and keep the answer in writing.

Do AI detectors work?

Much better than they did, and still not well enough to prove anything. The leading detectors are themselves AI: two of them are classifiers trained on human and AI writing, and the third scores how surprising the word choices are to a language model.65

Tests of 2023 found the detectors of that time "neither accurate nor reliable", and one found that, on average, they wrongly flagged most essays by non-native writers.6667 The leading ones of 2026 do much better on plain AI text. In a test of July 2026 by Epoch AI, a research institute, three leading commercial detectors caught nearly all the passages that current models wrote from a simple prompt. Two of them flagged none of 495 human passages, and the third flagged 19. When the models were asked to imitate a given author, about 13% of their passages went undetected, and about 26% in scientific writing.65 A working paper of 2025 also finds that the best commercial detector keeps false positives very low.46 The human passages in the 2026 test came from well-known writers, so it says little about a doctoral candidate's text or one written in a second language.65

More than half of 30 German university guidelines, in a German review of May 2026, call detectors unreliable and no firm basis for proof.68 Cornell's teaching centre says: "We currently do not recommend using automatic detection algorithms for academic integrity violations that flag the improper student use of AI." Caltech's deans "strongly discourage the use of GenAI tools designed to detect GenAI use in student writing".6970

I would rest nothing on a score, high or low. A test covers one set of texts at one moment, and a maker can update its detector without notice.65 What stands is your own record: your drafts, your version history and your log.

Do examiners run doctoral texts through detectors?

Some may. I found no rule on what doctoral examiners do, and no count of how often it happens. Basel's Faculty of Law and its Institute of Educational Sciences each have doctoral candidates sign a declaration that allows a software check of the dissertation; the text does not say whether that includes AI detectors.1071 In an interim case on an admission essay, a Munich court accepted a detection tool together with the examiners' own reading.28

I would assume a check is possible. Read the declaration you will sign, and ask your doctoral office what it checks.

A detector flagged my text, or someone suspects AI. What can I show?

What you actually have: your drafts, your version history, your log, and your account of what you did and what AI did. Do not tidy or rebuild a record after the fact. Yale's Graduate School asks for "the ability to provide a draft history" of work submitted, so keep yours from the start.60

Frankfurt's guidance for lecturers on exams, which does not mention doctoral candidates, requires the examiner to document the suspicion, says a detector alone is not enough, and fails the work only after a hearing if the student cannot refute it.72

Speak to your supervisor first, and ask whoever raised the doubt which rule applies.

I do not want to use AI at all. Will I fall behind?

Not in your doctorate, by any rule I found. None requires a doctoral candidate to use AI, and Oldenburg's doctoral recommendations say: "Not using generative AI must not be a disadvantage for doctoral candidates."73

Beyond the rules, this is much of what I teach, so weigh my view with that in mind. I believe that learning to work with AI now can give you an important advantage in your career, in research or outside it. I also believe that using AI should be a deliberate choice, made because you find it interesting and enjoy it for its own sake. Without that, in my view, it is counterproductive, and I would not advise it.

I would tell your supervisor, and if they have no view, your choice stands.

Data safety

What may go in

What should never go into an AI tool?

Three things, unless someone with authority over them has said otherwise: data about people, work you were sent to review, and material whose terms forbid it.

  • Data about people. Of 15 university documents, 10 forbid it and 3 allow it only with someone's permission.1 Stanford bars students from uploading "other individuals' personally identifiable information" to tools not approved for such data, and Yale says: "Assume all information shared will be made public."3916
  • Work you were sent to review. Of 9 university documents, 8 forbid reviewing with AI, as do all 6 publisher policies counted.1
  • Material whose terms forbid it. Licences can set limits.34

For data about people, ask your university's data-protection office. For a review, ask the editor or funder who sent it to you.

Is it safe to put my own unpublished draft into a chatbot?

It depends on the plan, and I would not call any plan safe as such. The EU's guidelines advise against uploading your own unpublished work "unless there are assurances that the data will not be re-used", without saying what counts as one.14 Harvard says not to enter confidential data, "including non-public research data", into publicly available tools. Columbia says researchers "must avoid uploading, or using as input, any unpublished research data". Chicago requires a security and privacy review before confidential data go into public tools.223374 Of ten vendors compared, nine let a personal or free tier feed improvement, or leave it to a setting.75 I would read your plan's terms, check the setting, and keep out anything about other people. Where the draft is yours alone and no rule of your institution or funder settles it, the decision is yours.

May I put a colleague's unpublished draft into a chatbot?

Not without asking them, in my view, and if you hold it as a reviewer, nearly every text forbids it. I found no rule that says whether your colleague's agreement settles the matter. Princeton's history department prohibits its graduate students from uploading "to any AI platform the unpublished work of others".40 Oxford warns that uploading published or unpublished work "may constitute a breach of copyright", a warning and not a ban.8

So ask your colleague, in writing, naming the tool. If the draft holds data about people, ask your data-protection office too.

Plans and terms

Does a paid plan, or switching off training, make a chatbot safe?

Not by itself, and neither one answers for data about people. A paid plan need not change training: OpenAI, the maker of ChatGPT, lists Plus and Pro with Free among the plans whose content it may use.76 Organisation plans at OpenAI, Anthropic (the maker of Claude), Google Workspace and Microsoft 365 are stated not to be trained on by default.76777879

OpenAI says that after you opt out, giving feedback means "the entire conversation associated with that feedback may be used to train our models".76 Feedback can be an exception on organisation plans too, as at Anthropic.77 I would switch training off for ordinary work, and ask your data-protection office about anything more.

What happens to what I type into a chatbot?

It depends on the tool and the plan, and no single period holds across vendors. These statements do not compare line by line:

  • OpenAI. A deleted chat is scheduled for permanent deletion "within 30 days", with exceptions for de-identified data and for security or legal duties.80
  • Google, consumer Gemini. Activity is deleted after 18 months by default. "A subset of chats are reviewed by human reviewers", and those are kept for up to three years, even if you delete your activity.81
  • Anthropic, consumer plans. Conversations you rate with thumbs up or down are kept for up to five years, even where you have not allowed training.82
  • DeepL, a translation tool, paid version. Texts "will not be permanently stored", and document translation runs in the European Economic Area.83

Read the page for your own plan. I would leave the feedback buttons alone on anything sensitive.

Where does a model's training data come from, and should that trouble me?

You decide what weight to give it, because I found almost no rule that asks you to. Johns Hopkins says: "The use of vast datasets to train AI models raises questions about data sourcing and consent."84 Germany's research funder, the DFG, recommends that its reviewers choose models with publicly viewable training data.85

The practical risk sits in the output: Cornell's task force warns that tools can repeat training data "with a level of verbatim accuracy that violates copyright",58 so I would check anything that reads like a quotation.

Is my university's approved tool really private?

It depends on which model sits behind it and on your institution's settings and contract. Approval does not tell you who sees what. GWDG, a German research computing centre, says "none of your chat histories or other user data is stored on our systems", a statement covering only the models it hosts.86

Ask your IT service or data-protection office which model answers, how long prompts stay, and whether it is cleared for personal data.

How do I find out which tools my institution has approved?

Look at your university's IT service and data-protection pages, then ask. I found no rule that sets one way of finding out.

Check what the approval covers, since cleared for non-sensitive data says nothing about interviews, and who it covers: Cologne's directive binds employees, including students employed there, and no other students.87 Ask your IT service, and keep the answer.

Data about people

May I put interview or participant data into an AI tool?

Usually not. Of 15 university documents on personal data and transcription, 10 forbid it and 3 allow it only with someone's permission; none allows it on declaration alone.1 I found no text that names interview data in commercial chat tools.

There is a path, and it is narrow, costly and complex. Your data-protection office and ethics committee check at least:

  • A legal basis, the ground in the law that allows you to use the data, such as consent or a task in the public interest, with a further condition if the data are sensitive, such as health, politics or religion.24
  • A contract with the provider, which your university signs, not you. The GDPR, the EU's data-protection law, allows only "processors providing sufficient guarantees".24 I would not rely on a personal plan.
  • A lawful route for sending data abroad. For a US company, one route is its listing under the EU-US framework, or the separate Swiss-US one for data leaving Switzerland.8889
  • A risk assessment (an impact assessment) where the risk is high.90
  • Information to your participants.

I would do without it: code by hand, or keep the data on your own machine, with your office's agreement.

What counts as personal data?

Any information about a person who can be identified, directly or indirectly, and a name need not be visible. That is the GDPR's definition, and Swiss law runs along the same line, federal for ETH and EPFL and cantonal at most other universities.2491 Whether a given detail points to someone is a question for your university's data-protection office. I would bring them a concrete example.

Is a transcript safe once the names are removed?

No, not by itself. German regulators say removing names and addresses is usually not enough.90 If you swap names for codes and keep a key list, the transcript is still personal data under the GDPR.24 Deleting the key does not make it anonymous if the content itself points to a person, so I would treat interview transcripts as personal data, and ask your data-protection office before you upload one. Programs that strike names on your own machine can do that first step, and the same limit applies.

Do my participants need to be told that AI will process their data?

Yes, as the texts stand, and that is a duty to inform, separate from consent. The GDPR lists what people must be told, with limited exceptions, and Swiss law requires this too, under whichever statute covers your university.2491

Consent is one legal basis among several, and the consent your ethics committee asks for is a separate matter; your data-protection office can tell you which basis applies.24 The EU's non-binding guidelines ask for the person's consent before their data go into an external AI system.14 I would name the tool and its purpose on the information sheet.

Must I tell my ethics committee or data-protection officer?

It depends on your university's rules and on the risk, and I would tell both before you start. Few texts say so outright. Oxford's research policy does: uploading personal data, "when done as part of undertaking research, requires ethics approval".8 Penn says review-board approvals "should be obtained prior to exposing research participant data to AI tools".51 Ask your ethics committee, and ask it whether the Swiss Human Research Act applies to your project, if you are in Switzerland.92

Are Zoom, Meet, cloud transcribers and dictation tools a data transfer?

It depends on where the audio goes. If it goes to the cloud, I would ask the same questions as for any other AI tool. I found no rule that names these tools.

Recording comes first. In Germany it is an offence, even for a participant, to record another person's non-public words without authorisation, and Switzerland punishes a participant's recording without the others' consent, on complaint.2526 Neither text mentions AI note-takers. The EU's guidelines ask you to tell people when such tools are in use and what the risks are.14 Ask your university's legal service or data-protection office, and I would ask everyone before you record.

Working without sending anything out

Can I use AI without sending anything out?

Yes, with a model that runs on your own machine, though today that is practical mainly for transcription. This changes where the data go, not your duties to the people in them, and your data-protection office's or ethics committee's permission may still be needed.

For interviews, two open-source programs made for qualitative research transcribe on your own computer and install like any other program:

  • aTrain, made by researchers at the University of Graz, from the Microsoft Store for Windows. Its makers say it "does not send recordings or transcriptions to the internet", and that on an ordinary business laptop a one-hour recording takes about three hours.93
  • noScribe, for Windows, Mac and Linux, by a sociologist at the Lucerne University of Applied Sciences, which "runs completely locally on your computer". Download it from its project page: a look-alike site sells transcription under its name.94

Both rest on Whisper, a speech-to-text model, and both tell the speakers apart. Check every transcript against the audio, because quality varies with the language and the recording.

A chatbot on your own computer is another matter. Programs such as LM Studio run one, and LM Studio says that when you run a model locally, none of your messages or documents leave your system.95 It takes a capable computer, by LM Studio's figures 16 GB of memory and 4 GB of graphics memory.95 In my view, a local chatbot that is practical for someone who is not an expert is still hard to put together today, so I would not start there alone. Ask your IT service whether your university runs models on its own servers (see Is my university's approved tool really private?).

These are the makers' own statements, so I would ask your IT service to confirm them before you rely on them.

In other roles

Can I use AI when I review a paper or a proposal?

For a paper, usually no: of 9 university documents that speak to reviewing, 8 forbid it, and so do all 6 publisher documents.1 For a proposal, funders differ. The DFG allows supporting use under its conditions, and the Humboldt Foundation only if nothing is processed or stored for another purpose.9650 Austria's research-integrity agency bars AI from the actual judgment.97 Columbia notes that the two large American research funders do not allow it: "NIH and NSF, among others, prohibit using Generative AI for peer review."33

Ask the journal's editor or the funder's office before you use anything, and write the judgment yourself.

Can I use AI to draft a grant proposal?

Usually yes, if you disclose what the funder asks and answer for the content. Of eight funder texts I analysed, not a census, none forbids it and four ask for disclosure.1 The DFG, Germany's research funder, wants AI use disclosed, except tools that do not affect the content.98 Where the call is silent, ask the funder's office.

Can I use AI in my teaching, for feedback or marking?

It depends on what you hand over. Bonn's guideline and Cologne's Philosophical Faculty bar fully automatic marking.9954 Only five texts I found touch this,6847100 too few to show a common position.

I would give the mark myself and tell your students what you use. Ask your university's data-protection office before you upload student work.

Who wrote this, and how

What I sell. I teach researchers and universities to work with AI, for a fee, and I built some of the tools I mention, the ai-log skill among them. Weigh my advice with that in mind.

Declaration of AI use. Drafted with Claude (Opus 5.5, in Claude Code) and written in my voice. AI helpers (Claude Sonnet 5.5) read the issuers' pages and sorted them on 4 and 5 November 2026, and Claude checked every quote against its source on 7 November 2026. I chose the scope, reviewed and corrected every answer, and take full responsibility for the content. No lawyer has read this page.

Corrections. If you find an error, tell me at contact@spetschinsky.com and I will correct it and date the change. What changed: first public version, 8 November 2026.

Sources

Consulted on 4 November 2026; the American universities on 5 November 2026; the 2026 detector test, the local programs and the changed vendor pages on 7 November 2026.

Notes (100)

  1. My own count, made with AI helpers (see Who wrote this, and how), of 205 published documents, from 12 Swiss universities, 16 German (the U15 group of research universities and TU Munich), 4 Austrian and 18 from the League of European Research Universities, 142 of them university texts, with publishers, societies and funders counted separately. It counts documents, not universities, so a university with several faculty rules counts several times. These are named lists, not a census; the sorting is a judgment, not re-checked mark by mark; 11 of the 50 universities had no text that could be read, which does not mean they have no rule. The counts cover student, teaching and research rules together. I found no published count of doctoral rules alone. ↩︎a ↩︎b ↩︎c ↩︎d ↩︎e ↩︎f ↩︎g ↩︎h ↩︎i ↩︎j ↩︎k ↩︎l ↩︎m ↩︎n ↩︎o ↩︎p ↩︎q ↩︎r ↩︎s ↩︎t ↩︎u ↩︎v

  2. Twelve American universities, chosen because readers know them, which is a different test from the named lists: Harvard, Yale, Princeton, Columbia, Stanford, MIT, Berkeley, Caltech, Chicago, Cornell, Penn and Johns Hopkins. They are examples, not a sample, and they are not in any count on this page. ↩︎

  3. Heidelberg University, Use of generative AI in doctoral dissertations, undated. Guidance. ↩︎a ↩︎b ↩︎c ↩︎d ↩︎e

  4. Harvard Griffin Graduate School of Arts and Sciences, Handbook for Students 2026-2027, "Use of Generative AI". Binding on its students. ↩︎a ↩︎b

  5. MIT, Department of Electrical Engineering and Computer Science, guidelines on generative AI in EECS theses (MEng, SM, PhD), effective 20 April 2023. A department's rule. ↩︎

  6. Cornell University, College of Agriculture and Life Sciences, Communication PhD programme, GenAI policy, undated. A programme's rule. ↩︎a ↩︎b ↩︎c ↩︎d

  7. University of Oxford, Faculty of Classics, Policy on acceptable use of AI, version 1, November 2025. A faculty's rule, binding on its students. ↩︎a ↩︎b

  8. University of Oxford, Policy for using generative AI in research. ↩︎a ↩︎b ↩︎c ↩︎d ↩︎e ↩︎f

  9. University of Innsbruck, Faculty of Catholic Theology, Good to know, valid from winter semester 2026/27. A faculty's rule. ↩︎a ↩︎b

  10. University of Basel, Faculty of Law, Lauterkeitserklärung zur Dissertation, 2024. A faculty's rule. ↩︎a ↩︎b ↩︎c

  11. University of Zurich, Empfehlungen zum Umgang mit generativer Künstlicher Intelligenz, undated. Guidance. ↩︎

  12. Cornell Graduate School, Use of generative AI in graduate admissions applications, updated 1 September 2026, the passage on enrolled students. ↩︎

  13. UC Berkeley Graduate Division, Resources for faculty and staff, undated. Guidance for instructors. ↩︎

  14. European Commission, Living guidelines on the responsible use of generative AI in research, fourth version, October 2026. Non-binding. ↩︎a ↩︎b ↩︎c ↩︎d ↩︎e ↩︎f

  15. SNSF, the Swiss research funder, Approach to AI in funding applications, 28 February 2024. ↩︎a ↩︎b

  16. Yale University, Provost and senior officers, Guidelines for the use of generative AI tools, 20 September 2023. Guidance. ↩︎a ↩︎b ↩︎c

  17. University of Basel, Research Ethics Committee, Guidelines on the responsible use of generative AI in research, version 1.5, December 2025. ↩︎

  18. MIT, AI and Education task force, report, 13 August 2026. A recommendation. ↩︎a ↩︎b ↩︎c ↩︎d

  19. University of Lausanne, Intégrité scientifique et usage responsable de l'IA générative, 8 July 2025. Recommendations. ↩︎

  20. MIT, Vice President for Research, The Use of AI in Research and Scholarship at MIT, last edited 27 August 2026. Interim guidance. ↩︎a ↩︎b

  21. EPFL, Tips for the use of generative AI in research and education, undated. ↩︎

  22. Harvard, Office of the Provost, Guidelines for using ChatGPT and other generative AI tools, undated. Initial guidelines, which say they are not new university policy. ↩︎a ↩︎b

  23. Swiss Federal Data Protection and Information Commissioner, AI and data protection, 24 September 2025. ↩︎

  24. European Union, General Data Protection Regulation, Regulation (EU) 2016/679: Article 4(1) and recital 26 on identifiability and anonymous data; Articles 7(3), 13 and 14 on consent and information; Article 28 on processors; Article 35 on impact assessments; Article 89(1) on research. ↩︎a ↩︎b ↩︎c ↩︎d ↩︎e ↩︎f ↩︎g

  25. German Criminal Code, section 201. ↩︎a ↩︎b

  26. Swiss Criminal Code, Article 179ter, SR 311.0, status 1 January 2024, not the latest consolidation. ↩︎a ↩︎b

  27. Verwaltungsgericht Kassel, press release 4/2026, 26 March 2026, on judgments of 25 February 2026, 7 K 2134/24.KS and 7 K 2515/25.KS. Appeal allowed; from the court's press release, not the judgments. ↩︎a ↩︎b ↩︎c

  28. Verwaltungsgericht München, decision of 8 May 2024, M 3 E 24.1136, interim relief. ↩︎a ↩︎b

  29. Austria, Hochschul-Qualitätssicherungsgesetz, section 2a (3) no. 2. ↩︎

  30. EU AI Act, Article 2 and recital 25, on the Commission's service desk. ↩︎

  31. Yale University Library, Generative AI for research: input, updated 29 October 2026. Guidance. ↩︎

  32. University of Helsinki, Use of generative AI in research, undated. Guidance. ↩︎a ↩︎b ↩︎c

  33. Columbia University, Office of the Provost, Generative AI policy, undated. Binding on the university's community. ↩︎a ↩︎b ↩︎c ↩︎d ↩︎e

  34. University of Fribourg library, AI restrictions for electronic resources, undated; covers ScienceDirect and Wiley Online Library. ↩︎a ↩︎b

  35. Xu et al., GhostCite, arXiv, 14 May 2026, Table III. A preprint; a small test in computer science. ↩︎

  36. Princeton University, Rights, Rules, Responsibilities, section 2.4, undated. Binding on its students. ↩︎a ↩︎b

  37. Princeton University, Scholarly Integrity, Disclosing generative AI, undated. Guidance, with a template. ↩︎a ↩︎b ↩︎c

  38. Wissenschaftsrat, press release 15/2026, 6 July 2026. Press release only, not the full recommendations. ↩︎

  39. Stanford University, Board on Conduct Affairs, Generative AI policy guidance, adopted 15 October 2026. The default for courses where an instructor sets nothing else. ↩︎a ↩︎b

  40. Princeton University, Department of History, Graduate student generative AI policy, undated. A department's rule. ↩︎a ↩︎b

  41. KU Leuven, Use of GenAI within the different phases of research, 3 September 2026. Guidance. ↩︎a ↩︎b

  42. University of Innsbruck, Faculty of Psychology and Sport Science, Richtlinien für die Verwendung von künstlicher Intelligenz, September 2025. A faculty's guidance for students. ↩︎a ↩︎b

  43. University of Dundee, Core policy on generative AI in education and student experience, updated 7 October 2026. A university policy for students. ↩︎a ↩︎b

  44. The Open University, Generative AI for students, updated 15 July 2025. Guidance. ↩︎

  45. ALLEA, The European Code of Conduct for Research Integrity, 2023, section 3.1. A code, not a statute. ↩︎a ↩︎b

  46. Jabarian and Imas, Artificial Writing and Automated Detection, NBER Working Paper 34223, September 2025. Abstract only. ↩︎a ↩︎b

  47. Leiden Law School, Use of genAI for research, undated. A faculty's guidance. ↩︎a ↩︎b

  48. University of Bern, IT Services, Generative KIs, undated. ↩︎

  49. Taylor & Francis and Routledge, AI Policy, undated. Journals may be stricter. ↩︎

  50. Alexander von Humboldt Foundation, Umgang mit generativer KI im Auswahlbereich, December 2025. Guidance. ↩︎a ↩︎b

  51. University of Pennsylvania, Guidance on the use of generative artificial intelligence, undated. Guidance. ↩︎a ↩︎b

  52. University of Zurich, Vetsuisse Faculty, Einsatz von generativer KI bei der Verfassung von Arbeiten, undated. A faculty's guidance. ↩︎a ↩︎b

  53. Elsevier, The use of generative AI in writing for Elsevier, July 2026. ↩︎

  54. University of Cologne, Philosophical Faculty, Leitlinie zur Nutzung von KI-Anwendungen in Studium und Lehre, 13 May 2026. A faculty's guidance, for coursework. ↩︎a ↩︎b

  55. Times Higher Education, Researchers not fully disclosing AI use, 11 September 2026. A newspaper's report of the survey, not the survey itself; its sampling is not given. ↩︎

  56. University of Bern, University Leadership, Richtlinien für den verantwortungsvollen Einsatz und die Nutzung von KI-Tools, 28 April 2026. Binding. ↩︎a ↩︎b

  57. University of Dundee, Guidance on proofreading of written submissions for assessment, updated 24 August 2021, before generative AI. ↩︎

  58. Cornell University, Research and Innovation, Generative AI in academic research, task-force report of December 2023. Guidance. ↩︎a ↩︎b

  59. Yale University Library, Generative AI for research: output, updated 29 October 2026. Guidance. ↩︎

  60. Yale Graduate School of Arts and Sciences, Academic integrity standards, 2026-27. ↩︎a ↩︎b

  61. Harvard, PhD Program in Biological and Biomedical Sciences, Graduate Student Policy, undated. A programme's rule. ↩︎

  62. University of Innsbruck, Faculty of Education, KI-Empfehlung für Studierende, December 2025. A faculty's recommendation. ↩︎

  63. Keenan, Smith and Seyfried, for the Scottish Graduate School of Social Science, Generative AI and the social science PhD, February 2026 by its file name, fieldwork May to November 2025. A self-selected UK survey, so it describes who answered. ↩︎

  64. University of Graz, Orientierungsrahmen zum Umgang mit generativen KI-Systemen in Studium und Lehre, version 2.0, 20 October 2026. Guidance. ↩︎

  65. Jaeho Lee, Epoch AI, AI detectors rarely flag human writing, but sometimes miss AI text imitating real authors, 15 July 2026. Pangram, GPTZero and Originality.ai, in their versions of June 2026, on 495 human passages of about 500 words written before 2022 and 594 passages by Claude Opus 4.8, GPT-5.5 and Gemini 3.1 Pro, with how each detector works as its maker describes it. A research institute's data insight, not peer-reviewed. ↩︎a ↩︎b ↩︎c ↩︎d

  66. Weber-Wulff and others, Testing of Detection Tools for AI-Generated Text, arXiv, 10 July 2023; International Journal for Educational Integrity 19, 26. Abstract only. ↩︎

  67. Liang and others, GPT detectors are biased against non-native English writers, arXiv, 10 July 2023; published in Patterns. A test set of TOEFL essays, not research writing. ↩︎

  68. Hochschulforum Digitalisierung, a German forum on digital higher education, Blickpunkt "Leitlinien-Check 2026", 19 May 2026. An analytical review of guideline texts, with nothing on doctoral rules. ↩︎a ↩︎b

  69. Cornell University, Center for Teaching Innovation, AI and academic integrity, undated. Guidance for lecturers. ↩︎

  70. Caltech, AI in teaching, updated 19 September 2025. Guidance. ↩︎

  71. University of Basel, Institute of Educational Sciences, Eigenständigkeitserklärung zur Einreichung einer Dissertation, valid from autumn semester 2023. A department's form. ↩︎

  72. Goethe University Frankfurt, Einsatz von generativer KI in der Lehre, updated 8 May 2026. Guidance for lecturers. ↩︎

  73. University of Oldenburg, Recommendations for the use of generative AI in doctoral studies, 12 November 2024. Recommendations. ↩︎

  74. University of Chicago, Generative AI guidance, undated. Guidance. ↩︎

  75. My own tally of one page per vendor. Nine: OpenAI, Anthropic, Google, Microsoft, Mistral, DeepL, Grammarly, Otter, and Wispr Flow, whose data controls, updated 18 August 2026, leave it to a setting for individual users. Zoom says it does not train on meeting content. Several plans were not covered. ↩︎

  76. OpenAI, How your data is used to improve model performance, updated about 20 October 2026. The vendor's own statement. ↩︎a ↩︎b ↩︎c

  77. Anthropic, Is my data used for model training on Claude for Work?, 18 August 2026. The vendor's own statement. ↩︎a ↩︎b

  78. Google, Generative AI in Google Workspace Privacy Hub, 6 October 2026. The vendor's own statement. ↩︎

  79. Microsoft, Data, privacy and security for Microsoft 365 Copilot, updated 3 November 2026. The vendor's own statement. ↩︎

  80. OpenAI, Chat and file retention in ChatGPT, updated 3 November 2026. The vendor's own statement. ↩︎

  81. Google, Gemini Apps Privacy Hub, 26 October 2026. The vendor's own statement. ↩︎

  82. Anthropic, Is my data used for model training?, 16 March 2026. The vendor's own statement. ↩︎

  83. DeepL, Privacy Policy, sections 4 and 7, undated in the part read. The vendor's own statement, for the paid version. ↩︎

  84. Johns Hopkins University, Generative AI tool guidance: guidelines, undated. Guidance for teaching. ↩︎

  85. DFG, Leitlinie zur Nutzung von Künstlicher Intelligenz in der Begutachtung, form 4.04, March 2026. For its reviewers; a recommendation on this point. ↩︎

  86. GWDG, Chat AI documentation, undated. The operator's own statement. ↩︎

  87. University of Cologne, Rektorat, KI-Richtlinie, 27 January 2026. A binding internal directive, not a statute. ↩︎

  88. European Commission, EU-US data transfers, undated; on adequacy decision 2023/1795, not consulted directly. ↩︎

  89. Swiss Federal Data Protection and Information Commissioner, New Swiss-US Data Privacy Framework, 15 August 2024. ↩︎

  90. Datenschutzkonferenz, the conference of Germany's data-protection authorities, Orientierungshilfe Künstliche Intelligenz und Datenschutz, version 1.0, 6 May 2024. Guidance. ↩︎a ↩︎b

  91. Switzerland, Federal Act on Data Protection, SR 235.1, Articles 5 and 19, status 1 September 2023. An English courtesy translation without legal force. ↩︎a ↩︎b

  92. Switzerland, Human Research Act, SR 810.30, Article 2, status 1 September 2023. An English courtesy translation without legal force. ↩︎

  93. Center for Data Science in Business and Society (BANDAS), University of Graz, aTrain, undated; the page says aTrain is not an official app of the university. The makers' own statement. ↩︎

  94. Kai Dröge, noScribe, project page, read on 7 November 2026, with the maker's affiliation and his warning about a look-alike site. The maker's own statement. ↩︎

  95. LM Studio, Privacy policy, effective June 2026, last updated 19 October 2026, and its system-requirements page, undated. The vendor's own statement; it also sells features that run in the cloud, which this does not cover. ↩︎a ↩︎b

  96. DFG, AI in the review process, undated. ↩︎

  97. OeAWI, the Austrian Agency for Research Integrity, Richtlinie zur guten wissenschaftlichen Praxis, adopted 18 June 2026. Self-regulation binding on its members. ↩︎

  98. DFG, AI in proposal preparation, undated. ↩︎

  99. University of Bonn, Leitlinie zum Umgang mit generativer Künstlicher Intelligenz im Bereich Studium und Lehre, 23 October 2025. Guidance, a framework for faculties. ↩︎

  100. TU Wien, Artificial intelligence in education, undated. Guidance. ↩︎

Have a question about Sergueï's work? Ask me.

Ask about my services

Answers come from an AI model and are general information: check your institution's own rules. Please don't paste unpublished work, personal data or anything confidential.

Hello! I can help you learn about Sergueï's AI services for academics. What would you like to know?