Verify and declare AI-assisted work
Can I defend this result and explain what the system contributed?
First confirm that the use was permitted, then verify the artefact, record the material contribution, and take responsibility for the final work.
In brief
- Permission, verification, declaration, and accountability are separate checks.
- No TUM-wide AI-declaration format is listed in the current public guidance; follow the instructions for the specific assessment.
- A detector score is not proof of authorship or misconduct.
Keep four questions separate
- Was the use permitted? The examiner determines permitted aids under the APSO, supplemented by the applicable FPSO and assessment instructions. A declaration cannot authorize a prohibited use.
- Does the result survive verification? Check factual claims, sources, quotations, calculations, data transformations, code, and assumptions using methods appropriate to each artefact.
- What did the system materially influence? Record the tool, version where available, purpose, affected parts, and how output was checked, changed, or rejected.
- Who accepts responsibility? You remain responsible for the submitted work. You should understand it, be able to explain it, and defend the decisions it contains.
Use the check that fits the artefact
Scroll horizontally to view all columns.
| Artefact | Check directly | Do not accept as proof |
|---|---|---|
| Factual claim or summary | Original source, scope, date, qualifications, and whether causal language was strengthened. | A second generated summary or a confident tone. |
| Citation or quotation | Existence, metadata, exact wording, location, context, correction status, and claim support. | A plausible DOI, quotation marks, or a source link alone. |
| Calculation or dataset | Definitions, units, denominator, exclusions, version, transformation, and independent reproduction. | A neat table or an unexplained result. |
| Code or configuration | Diff, dependencies, permissions, security implications, expected cases, edge cases, and execution. | Passing generated tests without reviewing what they cover. |
| Rule or policy statement | Currently effective official text, jurisdiction, actor, date, definitions, and exceptions. | A university blog, old translation, or general policy summary. |
Verify references outside the generated answer
- Search for the original record. Use the publisher, DOI registry, library catalogue, official database, or issuing organization rather than another chatbot response.
- Open the source. Confirm authors, title, date, venue, version, and whether the full text is the document that was described.
- Read around the evidence. Check the surrounding argument, method, population, uncertainty, and limitations. Topic overlap is not claim support.
- Check its current status. Look for corrections, retractions, superseding rules, later editions, or changed service documentation.
- Cite what you inspected. Reference the original evidence rather than the AI system that helped you discover or summarize it.
Keep a proportionate assistance record
The required form belongs to the assessment, programme, chair, journal, or supervisor. The fields below are a preparation record, not a universal TUM template.
- Tool and provider; model or version if the interface identifies it.
- Date and purpose of the use.
- Affected section, figure, analysis, code component, or work step.
- Material instructions or prompts where the applicable rule requires them.
- Output accepted, revised, or rejected and the substantive changes you made.
- Sources, tests, calculations, or human review used to verify the result.
- The exact assessment or publication rule followed.
At TUM, the concrete assessment rule controls
The Allgemeine Prüfungs- und Studienordnung (APSO) supplies the university-wide examination framework. Your degree's Fachprüfungs- und Studienordnung (FPSO) and the examiner's instructions add the requirements for your task. There is no blanket AI permission in the APSO.
The readable APSO consolidation (PDF, including the 12th amendment of 16 December 2025) was checked on 5 October 2026. Its cover says that the officially published text is binding. Find that text and your applicable FPSO in the central TUM statutes database.
- §12(5): the examiner determines permitted aids. In its written-examination context, including the other written work listed in §12(1), they are announced at least four weeks before the examination date.
- §18(9): the final-thesis declaration confirms independent authorship and lists sources and aids. This is not a form for every weekly exercise.
- §22(1): software may check unacknowledged copied passages or sources; the examiner decides whether deception occurred. A software result is not that decision.
- §25(3): after an examination result, a written, reasoned request can obtain inspection of the written work, related reports or oral-exam record. The request goes to the examination-board chair within one month of the result's announcement.
- §31: decisions affecting student rights require text form, reasons and instructions for challenging them.
Programme, School, chair, journal, and individual assessment instructions can add more specific requirements. Some ask for work-step documentation or prompts; others do not. Do not copy a local form and present it as TUM-wide.
Use an official template when it fits
The University Library's software-used template (PDF) records software, version and dates, purpose, affected pages or chapters, and a comment. It is a template for scientific work, not a mandatory form for every TUM assessment. Our recommended verification notes add your checks and changes; they are not a required column in that template. Follow your course or supervisor's required format first.
Do not substitute detection for evidence
A detector score does not establish who wrote a passage, whether a use was permitted, or whether deception occurred.
TUM’s current instructor decision guide states that reliable control of AI use is difficult to impossible. The APSO leaves the judgement about deception with the examiner. Fair process needs the applicable rule and concrete evidence, not an automated percentage alone.
- Do not upload your own or another person’s work to a public detector without an authorized data route.
- Do not interpret a detector score as the probability that misconduct occurred.
- Keep drafts, sources, methods, and decisions that can explain how the work developed.
- If your work is questioned, ask which rule and concrete evidence apply and request a fair opportunity to explain your process.
If your work is questioned
A request for explanation is not a final decision. Read the precise concern, the response route and the deadline. Preserve your actual drafts, source notes, code history and assistance record. Do not reconstruct a false history or submit a chatbot-written explanation that you cannot defend.
The Artemis plagiarism-case documentation shows one workflow where an instructor requests a student statement before deciding. Other assessments can use other procedures. This is not a university-wide promise of that interface or sequence.
Illustrative response preparation: suppose you are asked about a passage in Section 2. Locate the passage, its original source and the draft where you added it. Write a factual account of what you used, what you changed and what you checked. If your record reveals an error, identify it. Keep unrelated third-party or confidential content out of the response; ask the responsible contact how to provide relevant protected material.
For a formal decision, follow its challenge instructions and deadline. The inspection route above has a specific scope; it does not promise access to every document in every inquiry. §21's immediate written notice concerns defects in examination procedure, not every appeal. Ask the responsible examiner or examination board which route applies to your case. This guide cannot decide an individual examination dispute.
Before you submit
- I found and followed the rule for this assessment.
- I can explain every substantive claim, step, source, and decision.
- I checked citations, quotations, calculations, data, and code using suitable original evidence or tests.
- I removed or corrected output that did not survive verification.
- I declared material assistance in the form the assessment requires.
- I retained enough process evidence to reconstruct what I did.
- I did not disclose protected data or material to an unauthorized service.
Sources and scope
These sources support the guidance on this page. Product and institutional details can change; use the linked official information for current conditions.
- General Academic and Examination Regulations — readable German consolidation (opens in new tab)
Technical University of Munich · Through the 12th amendment, effective 1 January 2026
Readable consolidation; the officially promulgated German text controls and the applicable FPSO supplements it
Official TUM statutes and regulations database (opens in new tab)
- Using AI Wisely: Guidance for Students and Faculty (opens in new tab)
TUM ProLehre | Media and Didactics · Living guidance; checked 30 July 2026
TUM teaching and learning guidance; not legal advice
- TUM Citation Guide (opens in new tab)
TUM University Library · June 2026
TUM-wide advisory citation guidance; not examination law
- Fabrication and errors in bibliographic citations generated by ChatGPT (opens in new tab)
Scientific Reports · 7 September 2023
Historical model snapshots and one citation-generation task class
- Lateral reading: College students learn to critically evaluate internet sources (opens in new tab)
Harvard Kennedy School Misinformation Review · 23 February 2021
Website-evaluation intervention adapted here to source checking
- Living guidelines on the responsible use of generative AI in research (opens in new tab)
European Commission · Third version, May 2026
Non-binding research guidance; not a universal assessment template
- AI Attribution Guidelines (opens in new tab)
Cornell Center for Teaching Innovation · Living guidance; checked 30 July 2026
Adaptable teaching pattern, not TUM policy
- Wie entscheide ich, wofür ich den Einsatz von KI-Tools in meiner Lehre erlaube? (opens in new tab)
TUM ProLehre | Media and Didactics · Version 2, 9 March 2026
German teaching decision aid; not an examination statute
- Three categories of GenAI use in assessment (opens in new tab)
UCL Teaching & Learning · Living guidance; checked 30 July 2026
Communication pattern; explicitly not formal policy