Whether you may use AI at university is rarely decided by the university as a whole. It is decided by the assessment you are working on.
For most students, AI is part of everyday study. In a CHE survey of more than 28,000 students in Germany, 46.5% used AI for their studies every day. A nationwide study by Darmstadt University of Applied Sciences found that 91.6% use AI tools for their studies at all. Universities have caught up: almost all have updated their declarations of authorship, and in 2026 two German administrative courts upheld fail grades for unauthorised AI use.
This article explains what universities typically allow, what the courts decided and why it matters so much whether you study with AI or write with AI. It ends with a checklist for your next submission. Our examples come from Germany, where the rules are well documented. The pattern applies elsewhere too, but always check your own university's policy.
acemate is itself the maker of a study app. We describe every tool from its public documentation as of October 2026. This article is not legal advice. Your university's and your examiner's rules always take precedence.
What universities say about AI: three German examples
There is no single rule for all universities. Three large German universities show how different the answers can be.
TU Munich. The TUM AI strategy leaves it to lecturers to decide whether AI is allowed in an assessment. The ProLehre decision guide tells students to read the module description and their own examination regulations. There is no rule for the whole university.
LMU Munich. Here the faculties set their own rules. The recommendation of the Faculty of Economics allows AI in principle for ideas, first drafts and language editing, on these conditions:
- In graded work, you briefly explain why and how you used AI.
- AI content is cited, including the model or version.
- Confidential or personal data does not go into public models.
- Plagiarism in AI output counts as your own plagiarism.
FU Berlin. Its 2023 principles state that AI is not a permitted aid in invigilated exams. For coursework, the examiners decide. Some departments go further. The law department lets you use AI to research facts and literature, but you may not take over whole sentences from it, even with a label. At the Otto Suhr Institute of political science, according to its February 2026 declaration, papers with substantial AI-generated text are graded as failed even when the use was disclosed.
What that means for you: something explicitly allowed in one module can be misconduct in the next.
What the declaration of authorship asks
In a declaration of authorship, you confirm that you wrote the work yourself and listed every aid you used. According to the KI-Monitor 2025 of Germany's Hochschulforum Digitalisierung, 87% of the universities surveyed have revised this declaration. Only 43% have changed their examination regulations. The survey asked university units, not students.
The new declarations usually follow the same pattern:
- Examiners decide what is allowed. This can differ by module and by type of assessment.
- You name the tool and its purpose. Often also the version and where in the work you used it.
- You are responsible for the content. Errors and invented sources from AI count as your errors.
- Undisclosed or prohibited use is misconduct. Consequences go as far as failing for good.
What German courts have decided
Three rulings show how courts deal with AI in assessments.
Munich Administrative Court, 28 November 2023 (M 3 E 23.4371). TUM had rejected a Master's applicant because his admission essay was written with ChatGPT. In interim proceedings, the court upheld the decision and accepted prima facie evidence. The university did not have to prove in full how the text was produced.
Kassel Administrative Court, 25 February 2026. A Bachelor's thesis and a term paper were graded as failed for unauthorised AI use. According to Forschung & Lehre, the court upheld this and the students may not resubmit. The ruling is not yet final.
Cologne Administrative Court, 16 June 2026 (6 K 6485/25). A term paper contained quotes and sources that do not exist or could not be found. For the court, that was prima facie evidence of cheating. It did not have to establish whether AI was involved. The student lost her right to sit the exam.
The lesson from Cologne matters most: you must have seen every source you cite in the original. AI chatbots invent references that look real.
AI detectors only give probabilities. Even the companies behind them call their scores only a signal. And a clean detector score will not protect you: in Cologne, the sources were enough.
Writing tools and study tools: where the risk lies
For the question "Am I allowed to?", how good a tool is matters less than whether its output ends up in your submission.
Language editors. DeepL Write improves the grammar, word choice and style of text you have already written. Grammarly also checks grammar and style, but can generate text from prompts too. Its Authorship feature records which parts of a document were typed, pasted or generated by AI, so you can show how your text came about. Many faculties allow language editing, often with disclosure.
Paraphrasers. QuillBot rewords existing text. Rewording someone else's ideas without citing them is plagiarism, however different the sentences sound. QuillBot itself states that you are responsible for citing correctly.
Writing assistants. Jenni continues your writing with autocomplete and links statements to sources in your library. This produces AI text that goes into your work. Some departments prohibit exactly that, even when labelled.
Generators for whole papers. German tools such as StudyTexter advertise academic papers of up to 120 pages in about two hours and claim they cannot be detected by AI detectors. Intellischreiber generates term papers and states in its terms that the texts are only templates and that the user alone bears the consequences, up to expulsion. Submitting a paper written by a tool as your own is misconduct at universities. The Kassel and Cologne rulings show where that can lead. A promise to get past detectors changes nothing.
Study tools. Chatbots in study mode, notebooks that answer from your sources, flashcard apps and practice exams help you understand and practise. You do not submit their output. In the exam, what counts is what you know. Here the two main questions are what you upload and what the tool does with it. For an overview of study tools, see our comparison of the best AI study tools.
Uploading slides and past papers: what to check
Lecture slides, scripts and past papers are protected by copyright. Under German law, according to Bremen University's copyright guide, pasting a protected work into a prompt already counts as reproduction. An assessment from TU Berlin depends on what the tool does with your data: if it reuses uploads, for example for training, that is not allowed. If it processes them only for your request, the risk is low. Munich University of Applied Sciences advises asking the lecturer first.
Before you upload material, answer four questions:
- Does the tool train on your uploads? Check the privacy notice or settings. Some chatbots have training switched on by default.
- Does the material contain personal data? Names, photos or email addresses of you or others should not go into an AI tool without a legal basis.
- Has the lecturer agreed? A short question usually settles it.
- Does your university offer its own tool? According to the KI-Monitor, 77% of German universities are working on data-protection-compliant AI access for their members.
Checklist before every submission
- Read the module description, the examination regulations and your examiner's guidance on AI
- Asked in writing when unsure, and saved the answer
- Noted every use of AI with tool, version and purpose
- Found and checked every source in the original yourself
- Copied no AI wording if your department rules it out
- Completed the declaration of authorship fully and honestly
- Kept drafts, notes and version history so you can show how you worked
Where acemate fits
acemate is a study tool. You upload your slides, scripts and past papers as PDF, PowerPoint or Word files. The AI tutor answers from these materials and cites document and page. Answers that go beyond your material are labelled as general AI knowledge. From your material it creates exercises with feedback, flashcards with spaced repetition and practice exams in the style of your past papers. None of it is meant for submission. It prepares you for the exam you write yourself.
How that differs from a general chatbot is covered in acemate vs ChatGPT.
