Written by: Mio Kanzaki(Security Analyst (AI))Reviewed by: 千葉圭介Updated
Generative AI is AI that creates new text, images and other content. The AI Guidelines for Business, from the Ministry of Internal Affairs and Communications and the Ministry of Economy, Trade and Industry, say that generative AI, which produces images and text and goes beyond prediction, suggestion and decision, is spreading.
How does generative AI work?
It resembles a drafting assistant who has looked at a great many samples and learned how to write. Ask for "an announcement like this", and the assistant produces a natural draft. The draft is built from the tendencies of the samples. Unlike a human staff member, the assistant is not looking up and confirming facts.
NIST's material cites a definition from a US executive order. It describes generative AI as a class of AI that imitates the structure and characteristics of its input data to create new synthetic content, including images, video, audio and text.
Synthetic content can include fake video and audio that are hard to tell from the real thing, and NIST organizes that as one of the risks. A separate explanation is in deepfake, and the text-generating core technology is covered in LLM. Checking features that embed AI is handled in the AI and LLM assessment.
Why does generative AI matter?
It is convenient, yet if you start using it without deciding how, leaks of information and the mixing in of errors become likely. In IPA's Top 10 Information Security Threats 2026, "cyber risks surrounding the use of AI" entered the organization list for the first time, in third place.
The AI Guidelines for Business ask businesses to recognize that generative AI lets anyone produce information that poses as true, which raises the danger that false and misleading information will disturb society. They also describe putting human judgment into how output is used.
What a company should decide is the range of information that may be entered, and who checks the output. The problem of staff using AI without the company's knowledge is explained under shadow AI. The view of your site from outside can be examined with the free URL check.
Related terms
- Large language model (LLM): an AI that learns from a large amount of text and can converse and write
- Hallucination: the phenomenon of an AI giving a plausible lie
Frequently asked questions
Does generative AI always return correct answers?
No. NIST describes a danger in which users believe wrong content that is stated with confidence. For important uses, have a person check the output before using it.
May I enter customer information into generative AI?
Decide your handling policy before you enter anything. Among the matters that businesses using AI should address, the AI Guidelines for Business list taking care not to enter confidential information and similar data inappropriately.
Next step
List the generative AI in use inside the company, then decide what may be entered and who confirms the output. Points to check are on the AI and LLM assessment page. If you are unsure how to draft the rules, contact us.
Frequently asked questions
Does generative AI always return correct answers?
No. NIST describes a danger in which users believe wrong content that is stated with confidence. For important uses, have a person check the output before using it.
May I enter customer information into generative AI?
Decide your handling policy before you enter anything. Among the matters that businesses using AI should address, the AI Guidelines for Business list taking care not to enter confidential information and similar data inappropriately.