| Pseudonymous information |
Information that has undergone pseudonymization so that a specific individual cannot be recognized without additional information. It is still personal information. |
| Pseudonymization |
Processing that deletes or replaces part of personal information so that an individual cannot be recognized without additional information. |
| Indirect discrimination |
A case where applying the same criteria produces adverse results for a specific group and the justification for those criteria cannot be proven. |
| Audit trail |
A record maintained so that who did what, when, and why can be reconstructed later. |
| Developer business entity |
An AI business entity that develops and provides artificial intelligence (Article 2, Subparagraph 7, Item (a) of the AI Basic Act). |
| Personal information |
Information relating to a living individual that identifies an individual, information that can easily be combined with other information to identify an individual, and pseudonymous information. |
| Personal information controller |
A person who processes personal information directly or through another person in order to operate personal information files for business purposes. |
| Right to refuse |
A data subject's right to refuse an Automated Decision that has a material impact. Exceptions apply if the decision is made pursuant to consent, statutory provisions, or a contract (such as when the fact of the Automated Decision was clearly notified in advance). |
| False job advertisement |
An advertisement published to collect ideas or conduct publicity under the guise of hiring. Prohibited by Article 4 of the Fair Hiring Procedure Act. |
| Request for review |
A data subject submitting opinions (such as adding or correcting personal information) and requesting a review on whether they can be reflected in an Automated Decision. |
| Age Discrimination in Employment and Elderly Employment Promotion Act |
Act on Prohibition of Age Discrimination in Employment and Elderly Employment Promotion. |
| High-impact AI |
An AI system used in statutory areas (such as hiring and loan screening) that has a material impact or carries a risk of danger to life, bodily safety, or fundamental rights. |
| Unique identification information |
Resident registration number, passport number, driver's license number, and foreign resident registration number. |
| Disclosure obligation |
An obligation to publicly disclose the criteria and procedures of Automated Decisions and the methods of personal information processing so that data subjects can easily verify them. |
| Structured interview |
An interview where all applicants are asked the same job-related questions using the same scoring criteria. |
| Cross-border transfer |
Providing (including inquiry), entrusting the processing of, or storing personal information outside the country. |
| Worker Participation Act |
Act on the Promotion of Worker Participation and Cooperation. Prescribes the establishment and consultation matters of a labor-management council. |
| Basic screening materials |
A job seeker's application form, resume, and cover letter. |
| Equal Employment Opportunity Act |
Equal Employment Opportunity and Work-Family Balance Assistance Act. |
| Labor-management council |
A consultative body in which workers and employers participate. Established at the unit of a business or workplace with the authority to determine working conditions, with an exception for those with fewer than 30 regular employees. |
| Step-by-step analysis |
A method of calculating the pass rate for each group separately at every hiring stage to identify where disparities arise. |
| Proxy variable |
A variable that is not a protected attribute itself, but is strongly correlated with it and consequently reveals that attribute in its place. |
| Sensitive information |
Ideology, beliefs, labor union or political party membership, political views, health, sex life, etc., as well as genetic information, criminal records, biometric characteristics for identification purposes, and racial or ethnic information. |
| Use beyond the intended purpose |
Using personal information beyond the scope of the purpose and legal basis established at the time of collection. |
| Return claim period |
The period for requesting the return of hiring documents, determined and notified in advance by the recruiter within a range of 14 to 180 days from the date hiring is confirmed. Documents received via website or email are exempt from the return obligation. |
| Separate consent |
Consent obtained separately, distinguished from consent for other processing. |
| Protected attribute |
Characteristics that serve as the criteria for prohibition of discrimination. Age, gender, marital status, disability, region of origin, etc. |
| Unfair labor practice |
An employer's actions prohibited under Article 81 of the Trade Union and Labor Relations Adjustment Act, such as disadvantageous treatment on the grounds of union membership. |
| Adverse impact |
When a selection procedure results in a substantially lower selection rate for a specific group. |
| Meaningful human review |
Substantive review conducted by an individual with legitimate authority in a position to change the outcome. |
| Selection rate |
The proportion of applicants from a group who pass through to the next stage. |
| Selection rate ratio |
The value obtained by dividing the selection rate of a comparison group by the selection rate of the group with the highest selection rate. |
| Explainability |
The property of being able to represent the underlying mechanisms by which an AI system operates. |
| Right to request an explanation |
The right to request an explanation of the result, key criteria, and processing procedure regarding an Automated Decision. |
| Trustee |
A person who is entrusted with and processes personal information processing tasks. Includes sub-trustees. |
| In-depth screening materials |
Materials that demonstrate a job seeker's capabilities, such as portfolios and research publications. |
| Four-fifths rule |
A rule of thumb in U.S. federal guidelines that regards a selection rate ratio of less than 0.8 as an indicator of adverse impact. |
| Safety measures |
Technical, managerial, and physical measures to prevent leaks and other incidents. Internal management plans, retention of access logs, etc. |
| Affected person |
A person whose life, bodily safety, or fundamental rights are materially affected by an AI product or service. |
| Entrustment |
A personal information controller delegating the processing of personal information for its own business to a third party. |
| Deployer business entity |
An AI business entity that provides AI products or services by using artificial intelligence provided by a developer business entity (Article 2, Subparagraph 7, Item (b) of the AI Basic Act). |
| User |
A person who is provided with an AI product or AI service. |
| Anonymization |
Processing that makes an individual no longer identifiable using other information, even when reasonably considering time, cost, and technology. |
| Substantiating materials |
Materials that substantiate matters stated in basic screening materials, such as degrees, work experience, and qualification certificates. |
| Automation bias |
The tendency of humans to excessively trust and follow the outputs of automated systems. |
| Automated Decision |
A decision made by processing personal information through a fully automated system (including AI). Excludes automated dispositions by administrative agencies. |
| Affirmative action |
A provisional preferential measure for a specific gender to eliminate employment discrimination between men and women or to promote equal employment. |
| Reasonable accommodation |
Means and measures provided in consideration of disability characteristics so that persons with disabilities can participate on an equal basis with persons without disabilities. |
| Provision to a third party |
Transferring personal information so that the recipient may use it for their own purposes. |
| Restriction on the processing of resident registration numbers |
The principle that resident registration numbers cannot be processed even with consent unless there is a statutory basis, etc. (Article 24-2). |
| Material impact |
The degree to which an Automated Decision has a significant effect on the rights or obligations of a data subject, giving rise to the right to refuse. |
| Job analysis |
The task of systematically organizing the work actually performed in a job and the required knowledge, skills, and abilities. |
| Direct discrimination |
Treating someone differently on the grounds of a protected attribute. |
| Hiring documents |
Basic screening materials, substantiating materials, and in-depth screening materials. |
| Fair Hiring Procedure Act |
Act on Fair Hiring Procedure. Applies to hiring procedures of businesses (workplaces) with 30 or more regular employees; excludes hiring of public officials by the state or local governments. |
| Legal basis for processing |
The legal basis for processing personal information. Consent, statutory provisions, contract performance, etc. |
| Minimum collection |
The principle of collecting only the minimum personal information necessary for the purpose. The burden of proof rests on the controller. |
| Validity |
The property confirmed by objective evidence that a score or judgment is related to what it intends to measure (such as job performance). |
| Statistical significance |
The extent to which an observed difference is unlikely to have occurred by chance alone. |
| Interpretability |
The property of being able to understand what an AI output means in the context of its designed purpose. |
| Hallucination |
A phenomenon where AI plausibly fabricates content that is untrue or not present in the reference materials. |
| Restricting external knowledge |
A technique of instructing the model to use only the information in the provided documents and not general knowledge. |
| RAG (Retrieval-Augmented Generation) |
A method of retrieving documents relevant to a question and having the model answer based on their content. |
| Excessive Agency |
The risk where functions connected to an LLM possess functionality, permissions, or autonomy beyond what is necessary (OWASP LLM06). |
| Worker monitoring equipment |
Equipment or programs used to monitor workers at a workplace. Installation is a matter for consultation by the labor-management council (Article 20, Paragraph 1, Subparagraph 14 of the Worker Participation Act). |
| Miss rate |
The proportion of qualified applicants eliminated by AI (false negative rate). Compared across groups. |
| Prompt injection |
An attack that alters a model's behavior away from its intended manner through instructions in inputs or documents (OWASP LLM01). |
| Complete mediation |
The principle where downstream systems (HR database, permission verification APIs, etc.), rather than the LLM, enforce whether an action is permitted for every request. |
| GOVERN |
The function in the NIST AI RMF that cultivates a culture of risk management and establishes accountability. Cross-cuts across all other functions. |
| MAP |
The function in the NIST AI RMF that understands the context of an AI system to identify and frame risks. |
| MEASURE |
The function in the NIST AI RMF that quantitatively and qualitatively analyzes and evaluates risks. |
| MANAGE |
The function in the NIST AI RMF that allocates resources and responds to identified and measured risks. |
| NIST AI RMF |
The AI Risk Management Framework of the National Institute of Standards and Technology (NIST). |