Page 300 - CyberSecurityEssentialsEnglish

reasoning. Even so, sensitive data must be protected, and interviews, manuscripts, or
unpublished results must not be uploaded to public tools that the institution or the ethical
approval does not permit.

In research there is a difference between a source and an assistant. A source is material that

can be cited and inspected: an article, a report, data, a legal document, an interview, a
record, or a methodological note. A generative model is an intermediary that helps you work

with such materials, and it should not automatically be treated as a witness to the
correctness of their content. When a system says “studies indicate”, the first question is:
which studies? And did the tool actually read them, or did it construct a probable sentence?

International guidance proposes a human-centered approach in education and research,
with protection of privacy, ethical review, and transparency [1]. In scientific publishing,
editorial guidance affirms that human beings remain responsible for accuracy, originality,

and citations, and that AI tools are not listed as authors [12][13]. These principles are not
obstacles to productivity; they prevent time being lost correcting errors that could have
been avoided from the outset.

The best use is not always the most complex. Creating a simple evidence matrix, a reference

checklist, or a query log may be more valuable than building an autonomous research agent.
Design begins from the real bottleneck: is the problem finding the sources, reading them,

organizing them, comparing them, or writing? Then the tool is used at the point that raises
speed without hiding the origin of the information.

TABLE 7-1: TYPES OF AI USE IN RESEARCH AND THE LEVEL OF VERIFICATION

Type of use Example Level of verification

Proposing synonyms or sub- A quick logical review, then later
Exploratory
questions. testing.

Comparison with the original
Turning summaries into a table, or
Transformative material and retention of the
translating a text.
version.

Generating code or an initial Testing the code, the assumptions,
Analytical
coding of text. and the results on known data.

Direct verification of the source, the
Attributing a finding, a proportion,
Evidential method, and the context before
or a recommendation.
use.

Ranking policy options or a high- Explicit human approval and a
Decision support
impact choice. decision record.

299