AIAI Agent Dev Labaiagentdev-lab.com
Sign in to save progressSign in
Menu

G検定 · LESSON 7

Laws and contracts related to AI13%（当サイト設定） · 6 topics

Learn Japan's Act on the Protection of Personal Information, Copyright Act, Patent Act, Unfair Competition Prevention Act and Antimonopoly Act, AI development contracts and AI service contracts.

7.1Act on the Protection of Personal Information

Act on the Protection of Personal Information (personal information, individual identification codes, special care-required personal information, purpose of use, third-party provision, entrustment, anonymously and pseudonymously processed information, GDPR)

Concept

Personal information is information about a living individual that can identify a specific person (including when easily matched with other information to identify), plus individual identification codes (face recognition data, My Number, passport numbers etc.). Acquiring special care-required personal information (race, beliefs, medical history, criminal record etc.) in principle requires consent. Specify the purpose of use as far as possible; use beyond it in principle requires consent. Third-party provision in principle requires consent, but entrustment, business succession and joint use aren't third parties. Anonymously processed information is processed so individuals can't be identified or restored; pseudonymously processed information can't identify individuals unless matched with other information and is mainly for internal analysis. The EU GDPR is a regulation that can also apply to businesses outside the EU.

How it works / steps

  1. Judge whether the data handled is personal or special care-required personal information
  2. Specify the purpose of use and notify or publish it
  3. When giving data to an outside business, sort out whether it is entrustment or third-party provision

Workplace example

When giving customer data to an outside developer for AI development, treat it as entrustment, supervise security management, and don't allow the developer to use it for its own purposes.

Common confusions and exam tips

Mini quiz

Q1. Consider the categories of information under Japan's Act on the Protection of Personal Information (APPI). Which category covers information processed so no specific individual can be identified or restored, which can be given to third parties without consent under certain conditions?

  1. Anonymously processed information
  2. Personal information containing an individual identification code
  3. Pseudonymously processed information
  4. Special care-required personal information
Show answer and explanation

Answer: A Anonymously processed information

Creating anonymously processed information carries duties such as public disclosure. (Deciding point: neither identifiable nor restorable; third-party provision allowed.)

Q2. Consider the categories of information under Japan's Act on the Protection of Personal Information (APPI). Which category covers information processed, such as replacing names with IDs, so individuals can't be identified without matching other information, used for internal analysis but in principle not providable to third parties?

  1. Personal information containing an individual identification code
  2. Anonymously processed information
  3. Pseudonymously processed information
  4. Special care-required personal information
Show answer and explanation

Answer: C Pseudonymously processed information

Pseudonymously processed information makes changing the purpose of use easier but in principle bans third-party provision. (Deciding point: unidentifiable without matching, for internal analysis.)

Reference links

7.2Copyright Act

Copyright Act (works and creativity, Article 30-4, AI-generated content, terms of use)

Concept

A copyrighted work is a creative expression of thoughts or sentiments; ideas and data themselves aren't protected. Article 30-4 of the Copyright Act in principle allows, without the right holder's permission, uses not aimed at enjoying the thoughts or sentiments expressed in a work, such as information analysis (AI training etc.). Cases that unduly harm the copyright holder's interests are excluded. AI-generated content is unlikely to be recognized as a work without human creative contribution, and can infringe if similar to and reliant on an existing work. Use of trained models and services also follows their terms of use.

How it works / steps

  1. Check whether the data is a copyrighted work
  2. Check whether training use is within Article 30-4 (no purpose of enjoyment included)
  3. Check similarity to existing works when using generated content

Workplace example

Using articles collected from the web for AI training can be allowed under Article 30-4, but having the AI output a specific author's works and selling them carries infringement risk.

Common confusions and exam tips

Mini quiz

Q1. Consider AI and copyright (Japanese law). Selling an AI-generated image found to resemble an existing character image and to rely on that work. What is the most appropriate view?

  1. It is not a copyrighted work
  2. It can be a copyrighted work if there is human creative contribution
  3. Under Article 30-4 of the Copyright Act, it can in principle be used without permission
  4. It can be copyright infringement
Show answer and explanation

Answer: D It can be copyright infringement

Even with generative AI, similarity and reliance can make it copyright infringement. (Deciding point: similarity and reliance exist.)

Q2. Consider AI and copyright (Japanese law). A work that a person finished by refining prompts many times, choosing among many outputs and adding edits. What is the most appropriate view?

  1. Under Article 30-4 of the Copyright Act, it can in principle be used without permission
  2. It can be copyright infringement
  3. It is not a copyrighted work
  4. It can be a copyrighted work if there is human creative contribution
Show answer and explanation

Answer: D It can be a copyrighted work if there is human creative contribution

If a person has creative intent and makes creative contributions such as specific instructions, selection and editing, it can be recognized as a copyrighted work. (Deciding point: human creative intent and contribution.)

Reference links

7.3Patent Act

Patent Act (inventions, novelty, inventive step, employee inventions)

Concept

An invention under the Patent Act is a highly advanced creation of technical ideas using laws of nature. To obtain a patent it must meet requirements such as industrial applicability, novelty (not publicly known before filing) and inventive step (not easily conceived by an expert in the field). In AI, trained models and methods or devices using AI can be protected as inventions, while AI itself is held unable to be an inventor. Employee inventions are inventions made by employees as part of their duties; by contract or rules the company can acquire the right to obtain a patent, and the employee has the right to reasonable benefits.

How it works / steps

  1. Consider filing before publishing an invention (to avoid losing novelty)
  2. Sort out the scope of AI-based inventions (model, method, device)
  3. Put employee invention rules in place

Workplace example

File a patent for your in-house image inspection AI mechanism before presenting it at an academic conference, securing novelty.

Common confusions and exam tips

Mini quiz

Q1. Which best describes how copyright arises?

  1. It arises automatically at creation, with no registration or procedure needed
  2. It arises when the work is sold at a profit and not for free works
  3. It first arises on registration with the Agency for Cultural Affairs; unregistered works aren't protected
  4. It arises when an application to the Patent Office is examined and registered
Show answer and explanation

Answer: A It arises automatically at creation, with no registration or procedure needed

Copyright arises without formalities. Patent rights require application, examination and registration.

Q2. Consider how to protect intellectual property. You want to protect the expression of program source code written by your engineers from unauthorized copying by others (no procedure needed). Which protection is most appropriate?

  1. Trade secret (Unfair Competition Prevention Act)
  2. Copyright
  3. Patent right
  4. Shared data with limited access (Unfair Competition Prevention Act)
Show answer and explanation

Answer: B Copyright

Programs are protected as copyrighted works. Ideas and methods themselves are not. (Deciding point: the program's expression, no formalities.)

Reference links

7.4Unfair Competition Prevention Act and Antimonopoly Act

Unfair Competition Prevention Act (three requirements for trade secrets, shared data with limited access) and Antimonopoly Act

Concept

Protection as a trade secret under the Unfair Competition Prevention Act needs three requirements: secrecy management (managed as secret), usefulness (useful to business) and non-public status (not publicly known). Shared data with limited access is large volumes of data managed with IDs and passwords and provided to specific parties (managed electronically), protected from improper acquisition and use. The Antimonopoly Act regulates acts that hinder fair competition, such as data hoarding and abuse of a superior bargaining position.

How it works / steps

  1. Restrict access to important data and mark it confidential to secure secrecy management
  2. Manage data provided externally as shared data with limited access through contracts and ID management
  3. Check that requests for data from business partners aren't an abuse of a superior bargaining position

Workplace example

Mark training data “confidential,” limit the employees who can access it and keep logs, so it is protected as a trade secret.

Common confusions and exam tips

Mini quiz

Q1. Consider how to protect intellectual property. You want to protect large volumes of valuable but non-secret weather data, provided to member companies via ID and password, from improper acquisition. Which protection is most appropriate?

  1. Copyright
  2. Shared data with limited access (Unfair Competition Prevention Act)
  3. Database work
  4. Trade secret (Unfair Competition Prevention Act)
Show answer and explanation

Answer: B Shared data with limited access (Unfair Competition Prevention Act)

Shared data with limited access was introduced by a 2018 amendment and covers data not managed as secret. (Deciding point: data provided to specific parties under electronic management.) [Other options] Trade secrets must be managed as secret. Weather data itself is fact, not a work, and a database work needs creativity in selecting or organizing information.

Q2. Which is the correct set of three requirements for protection as a trade secret under the Unfair Competition Prevention Act?

  1. Creativity, expressiveness, originality
  2. Novelty, inventive step, industrial applicability
  3. Secrecy management, novelty, creativity
  4. Secrecy management, usefulness, non-public status
Show answer and explanation

Answer: D Secrecy management, usefulness, non-public status

Novelty and inventive step are patent requirements; creativity is a requirement for copyrighted works.

Reference links

7.5AI development contracts

AI development contracts (Contract Guidelines on Utilization of AI and Data, exploratory phased approach, NDA, quasi-mandate and contracts for work)

Concept

METI's “Contract Guidelines on Utilization of AI and Data” note that AI development results are hard to guarantee in advance, and present an “exploratory phased” approach contracting per stage: assessment → PoC → development → additional training. Sign an NDA (non-disclosure agreement) first. A quasi-mandate contract pays for performing work without a duty to complete; a contract for work pays for completing the work. In AI development, because performance is uncertain, a quasi-mandate is said to suit many stages. Also set rights and terms of use for trained models and generated outputs in the contract.

How it works / steps

  1. Contract per stage, deciding the purpose and deliverables of each
  2. Consider a quasi-mandate for highly uncertain stages
  3. Set rights and scope of use for training data, trained models and deliverables

Workplace example

In developing image inspection AI, first run a PoC under a quasi-mandate contract, and sign the development-stage contract once the target accuracy is confirmed.

Common confusions and exam tips

Mini quiz

Q1. You are choosing a contract type for AI development. At the assessment stage, before disclosing data, you want the other party to promise not to use the information beyond its purpose or leak it. Which fits best?

  1. Maintenance contract
  2. Quasi-mandate contract
  3. Contract for work (ukeoi)
  4. Non-disclosure agreement (NDA)
Show answer and explanation

Answer: D Non-disclosure agreement (NDA)

Sign an NDA before disclosure, defining the scope, purpose and period of confidential information. (Deciding point: handling of disclosed information.)

Q2. You are choosing a contract type for AI development. After production operation starts, you want ongoing incident response and maintenance of model performance. Which fits best?

  1. Maintenance contract
  2. Quasi-mandate contract
  3. Contract for work (ukeoi)
  4. Non-disclosure agreement (NDA)
Show answer and explanation

Answer: A Maintenance contract

A maintenance contract defines the scope of support and how retraining is handled. (Deciding point: ongoing post-launch support.)

Reference links

7.6AI service contracts

AI service contracts (SaaS, terms of use, data usage rights, accuracy guarantees)

Concept

When providing AI services via SaaS or similar, the terms of use or contract set rights to input data and outputs, whether the provider may use users' data for training or service improvement (data usage rights), whether accuracy is guaranteed, the scope of liability (disclaimers), and conditions for suspending or changing the service. Because AI output is probabilistic, terms often state that accuracy is not guaranteed and that users are responsible for checking.

How it works / steps

  1. Check and state whether the provider uses input data for training
  2. Set that accuracy isn't guaranteed and the scope of liability
  3. Set rights to outputs and terms of use

Workplace example

A business AI translation service states in its terms that customer input data isn't used for training and that translation accuracy isn't guaranteed.

Common confusions and exam tips

Mini quiz

Q1. A Japanese company offers an online service to individuals in the EU. How does the GDPR apply?

  1. Japan's adequacy decision means Japanese companies have no GDPR obligations
  2. It doesn't apply to Japanese companies without an EU establishment
  3. It can apply to non-EU businesses offering goods or services to individuals in the EU
  4. Transfers of personal data outside the EU are prohibited without exception
Show answer and explanation

Answer: C It can apply to non-EU businesses offering goods or services to individuals in the EU

The GDPR has extraterritorial reach. Japan has an EU adequacy decision, so personal data can be transferred under certain conditions. [Other options] The GDPR applies extraterritorially. The adequacy decision allows EU-to-Japan transfers and doesn't remove obligations. Transfers are possible with adequacy or appropriate safeguards.

Q2. Which is the most appropriate license check when using source code or data for AI development?

  1. Violating license terms is fine if you rewrite the code a little
  2. Code published online may be freely used commercially regardless of license
  3. Check and follow the terms of licenses and terms of use (commercial use, attribution duties, etc.)
  4. Licenses matter only for personal use, not for developing company products
Show answer and explanation

Answer: C Check and follow the terms of licenses and terms of use (commercial use, attribution duties, etc.)

A license is the condition under which the copyright holder permits use. Some, like the GPL, require derivatives to be published.

Reference links

Lessons are written by this site based on public information and do not guarantee exam content. Product specifications change, so check the latest information at the reference links.

Notes (unofficial, original questions, no pass guarantee)