A rule about responsibility, not about tools
South Korea’s Ministry of Science and ICT has finalised an AI research ethics guide for national research and development, after review and approval by the steering committee of the National Science and Technology Advisory Council. The governing principle is narrow and clear: AI’s role in publicly funded research is limited to that of a tool, and responsibility and rights for the final output rest with the researcher.
That is a position, not a truism. It rules out the defence that a model produced the error, and it keeps authorship and accountability attached to a person.
Seven requirements
The guide sets out seven, according to Digital Today. Researchers must verify AI-generated errors, gaps and source reliability themselves. They must review AI output against their own expertise rather than accepting it. They must disclose the model, its version, the purpose and the scope of use.
They must be able to explain how an AI-assisted result was produced. They must follow their institution’s own rules. They must not enter personal information into external AI services and should use tools their institution has approved. And evaluators must assess submitted material without using external AI services at all.
The disclosure rule is the load-bearing one
Six of the seven restate professional judgement in a new context. The third does something the others do not: it makes the model and version part of the record.
That matters because a result produced with one version of a model is not reproducible with another, and because the failure modes the ministry lists — inaccurate information, replicated bias, leaked personal or security data, unclear citation — are version-specific. Without the disclosure, a reader cannot tell which of those risks applied.
The evaluation ban
The seventh rule is the strictest. Reviewers of national R&D submissions may not put the material through an external AI service. The concern is not only judgement but confidentiality: an unpublished proposal pasted into a commercial chatbot has left the institution.
What comes next
The ministry plans to distribute the guide and explanatory material to research institutions, provide self-assessment checklists, and create disclosure forms for AI use in R&D reports. The form is where this becomes real. A principle in a guidance document is advisory; a field on a mandatory report is a record that can be audited, and it is the point at which a researcher has to commit in writing to what the machine did.
