South Korea tells funded researchers to disclose AI, bars hidden prompts
South Korea's Ministry of Science and ICT has issued national guidance on how researchers working on government-funded projects should use AI. The National R&D AI Research Ethics Guide was announced in a ministry press release dated 20 September 2026, after approval by the steering committee of the National Science and Technology Advisory Council. It sets one basic principle, that AI is a tool and the researcher is responsible for the result, and makes seven recommendations. It also names a specific kind of cheating: hiding instructions in a document so that an AI system used in reviewing it gives a better assessment.
Summary
- The guide is aimed at all researchers who use AI in national research and development (R&D), meaning research funded by the South Korean government. It also sets out some duties for their institutions.
- Researchers are asked to disclose the AI model and version, and the purpose and scope of use, when they submit or present their final output. A sample disclosure form and a 10-item self-check list are included.
- The guide says attempts by people being evaluated to distort an assessment using AI, such as hidden prompts, are research misconduct and may be sanctioned under existing law.
Key takeaways
- The guide states that it is not intended to regulate AI use. Its recommendations are written as things researchers should do, and it sets no penalties of its own.
- The one place it mentions sanctions is hidden prompts, and there it points to existing law without naming the statute or the penalty.
- The press release gives no date from which researchers are expected to follow the guide, and says the ministry will keep revising it after feedback from research institutions.
What was announced
The source is a 13-page press release with the full guide attached, issued by the Ministry of Science and ICT for release on 20 September 2026 and archived by the Korea Development Institute's economic information centre. The release says the guide was confirmed after deliberation and a vote by the steering committee of the National Science and Technology Advisory Council. The ministry's Research Ethics and Rights Protection Division is listed as the responsible office, with the Korea Institute of S&T Evaluation and Planning (KISTEP) named as the related agency. The document is in Korean, and the translations in this article are ours unless stated otherwise.
The ministry's stated reason is that AI is now used throughout the research process and brings new risks alongside higher productivity. The release lists inaccurate information, the repetition of bias in existing material, leaks of confidential and personal data, and improper citation of material whose source is unknown.
As background, the guide cites two 2025 surveys of researchers. It says a Wiley survey found the share of researchers who had used generative AI in any part of their work rose from 57% to 84% in a year, and that an Elsevier survey found the share using AI for work rose from 37% to 58%. We have reported these figures as the guide gives them and have not checked them against the original surveys.
The recommendations are written for generative AI based on large language models (LLMs), the type of system behind chatbots such as ChatGPT. The guide says they also apply, where relevant, to AI models an institution has built or trained itself and to agentic AI, meaning systems that carry out multi-step tasks with less direct instruction.
The principle and the seven recommendations
The basic principle is that in national R&D the role of AI is limited to that of a tool, and that responsibility and rights over the final output belong to the researcher who used it. Two points follow from it in the guide. AI should be confined to a supporting role that makes research more efficient, with examples including summarising literature, translation, grammar correction, basic coding, organising data and simulation. And AI cannot hold the status or rights of an author or inventor, because it cannot bear legal or moral duties.
| Recommendation | What the guide asks |
|---|---|
| 1. Verify the facts | Check AI output for errors, missing key material, and whether cited sources exist and are genuine. |
| 2. Judge for yourself | Review AI-generated material against your own expertise and its logical soundness, and rework it to reflect your own view. |
| 3. Be transparent | Disclose the model and version, and the purpose and scope of use, when submitting or presenting the final output. |
| 4. Manage reliability | Be able to explain logically how the research result was produced, even where AI was used. |
| 5. Follow policies and rules | Check and follow the law and your institution's rules. Before submitting a paper, check the journal or publisher's AI policy, since these differ. |
| 6. Keep research secure | For research that needs security, use only tools inside an environment your institution has approved. Institutions should provide that environment. |
| 7. Protect personal data | Take care not to enter research data containing personal information into AI services. Where it cannot be avoided, de-identify the data first. |
The guide gives examples of the uses that should be disclosed under the third recommendation: gathering information, drawing inferences from data, organising research data, checking grammar and spelling, and translation. It says the purpose of disclosure is to make results more trustworthy and to make clear that the researcher is the person finally responsible.
The disclosure form and the self-check list
Two sample documents are attached to the guide. The first is an AI use disclosure form for reports and other outputs, which the guide says institutions may adapt. It has four parts.
| Part of the form | What goes in it |
|---|---|
| Basic information | Project or paper title, principal investigator or author, and the dates of the research |
| Tool and model | The name of the tool and model. The worked example lists two commercial chatbots, one with a model version |
| How it was used | A description by task. The example covers finding and sorting prior studies, producing a diagram of the method, and correcting English in the abstract and conclusion |
| Period of use and confirmation | The months of use, then a signed statement that the author is responsible for the core ideas, data, analysis, conclusions and final manuscript |
The second is a 10-item self-check list, to be read before the research starts and completed before the final output is submitted. Its items include checking that no false references, altered data or distorted images are included; confirming that the researcher carried out the core analysis and judgement personally; considering whether results obtained with AI can be reproduced under the same conditions; not uploading unpublished data, raw data or technology awaiting a patent application to commercial AI services; and de-identifying personal and sensitive information before entering it.
What the ministry said, and what is not yet known
In the release, Hong Soon-jung, director general of the ministry's Performance Evaluation Policy Bureau, said AI is a useful tool for research efficiency but that final responsibility always lies with the researcher. In the English translation published by DigitalToday, Hong said: "I hope this guide will serve as an opportunity to establish a culture of responsible AI use in the research field and ensure the integrity of national R&D research."
The ministry said it would send the guide and an explanatory booklet to research institutions, post them on its website and on KISTEP's, and continue to revise the guide in response to feedback. The Korean daily Financial News reported on 28 September on what the guide means for research institutions in Jeju, including those whose work involves confidential company data and health information.
- Not stated: a start date, or whether funding agencies will require the disclosure form with project reports. The guide describes the form as an example.
- Not stated: which laws and which sanctions apply to hidden prompts.
- Not seen by us: the explanatory booklet mentioned in the release. We read the press release and the attached guide.
- Not addressed: how compliance will be checked. The guide does not mention AI detection software.
What this means for students and researchers
- Research students in South Korea: if your work is funded through a national R&D project, expect your institution to ask which AI tools you used, for what and when. Keeping a simple log as you go is easier than reconstructing it at the end.
- Anyone submitting a paper: the guide repeats a point that applies everywhere. AI policies differ between journals and publishers, so check the one you are submitting to before you submit.
- Anyone asked to review: do not paste a manuscript, proposal or report into an external AI tool unless the journal or funder allows it. The guide treats this as a confidentiality problem first.
- Students elsewhere: the disclosure form is a useful model for an AI use statement in a thesis or assignment where your institution asks for one: tool and version, what it was used for, when, and a statement that the work is yours.
Sources
- Ministry of Science and ICT (South Korea), National R&D AI Research Ethics Guide: press release and attached guide, 13 pages, in Korean, for release 20 September 2026. Copy held by the KDI Economic Education and Information Center. The release says the guide is also posted at msit.go.kr and kistep.re.kr.
- Jin-ho Lee, South Korea sets AI research ethics guidelines for national R&D, final responsibility rests with researchers, DigitalToday, 20 September 2026.
- Jung Yong-bok, report on the National R&D AI Research Ethics Guide, Financial News, 28 September 2026 (in Korean).
- Nikkei Asia, 'Positive review only': Researchers hide AI prompts in papers, 1 July 2025.
- Retraction Watch, Weekend reads, 26 September 2026 (lists the South Korean guidelines among the week's research integrity items).
FAQ
What is South Korea's National R&D AI Research Ethics Guide?
It is guidance from the Ministry of Science and ICT, announced on 20 September 2026, on how researchers working on government-funded research should use AI. It sets one basic principle, makes seven recommendations and includes a sample disclosure form and a self-check list.
Is the guide legally binding?
The guide says it is not intended to regulate AI use, and its seven recommendations carry no penalties of their own. The exception is its statement that hidden prompts and similar attempts to distort an evaluation are research misconduct that may be sanctioned under existing law.
What do researchers have to disclose?
The guide asks them to disclose the AI model and version and the purpose and scope of use when they submit or present their final output. The sample form also asks for the period of use and a signed statement that the researcher is responsible for the work.
What is a hidden prompt?
It is an instruction placed in a document where a human reader will not see it, for example in white text, written for an AI system that may be used to assess the document. Nikkei Asia reported in July 2025 that it had found such prompts in 17 preprints on arXiv.
Can reviewers use AI under the guide?
The guide says evaluators should not enter material under review into external AI, and should base their assessment on their own expertise and judgement. It says accepting AI output without the evaluator's own review undermines the purpose of expert evaluation.
What happens next
The guide gives researchers on South Korean state-funded projects a single national reference for AI use, and it names hidden prompts as misconduct. How much it changes practice will depend on what funding agencies and institutions do with the sample forms, which the release does not say. We will update this article if the ministry publishes an implementation date, the explanatory booklet or a revised version.
How AI detectors work
A plain-English guide to the methods behind AI-writing detectors, and what a result can and cannot tell you.
