AI Services and Personal Information Protection: What Companies Should Check First

Hello, I'm Noriaki Asato, Representative Attorney at LegalAgent.

Protecting personal information is an issue to check both for companies that develop and provide AI services themselves and for companies that introduce AI internally to streamline their operations. Generative AI processes a wide variety of data, such as text, audio and images. Where that data includes personal information, it is essential to confirm consistency with the Act on the Protection of Personal Information (APPI).

In legal work involving AI, the question is not only what technology is used but whether you can explain what data you handle, for what purpose and under whose management responsibility. When offering services to businesses, there are increasingly situations in which the legal department of the customer company asks for a detailed explanation.

Note that, following the so-called triennial review, an amendment act was enacted on July 10, 2026, including the creation of an administrative surcharge system, a special rule on consent for use in preparing statistics and the like (including AI development) that meets prescribed requirements, and rules on handling personal information of persons under the age of 16 (Commentary on the 2026 Amendment to the APPI). The amendment act was promulgated on July 17. According to the announcement by the Personal Information Protection Commission (Japanese), it will, with some exceptions, take effect on a date specified by Cabinet Order within two years of promulgation. We will put practices in order by clearly distinguishing enactment and promulgation from entry into force, and by taking into account the effective date of each provision and the specific rules set out in Cabinet Orders and ministerial ordinances.

Starting Point: Confirming the Data Flow

In establishing practices for AI and personal information protection, the first thing to confirm is the flow of data. Trace what information users enter on which screen, and whether that data is stored on your company's servers, transferred to an external AI API, saved as communication logs, or used for machine learning of a model.

If you write a privacy policy while this data flow remains unclear, discrepancies may arise between actual practice and what the policy states. In particular, in setups that incorporate a generative AI API provided by another company, the input information is sent to the external provider's servers. Check the contractual relationship between your company and the external provider, the scope of the data processing entrusted, the retention period for logs and so on.

As the person in charge of legal affairs, have the product division or development team share the data flow in diagrams or tables. Having materials that let you confirm the routes also makes it easier to make the legal distinction between entrustment and third-party provision discussed below.

Specifying the Purpose of Use and How to Describe It

When handling personal information, specify the purpose of use as concretely as possible. In services that use AI, there is a tendency for the description to become abstract. Boilerplate wording alone, such as "to improve our services" or "to improve convenience," may not fully explain the actual data processing.

For example, explain things in line with the actual processing, such as providing a function that summarizes text entered by users, automatically generating candidate answers in customer support, managing information on job candidates, or improving the UI by analyzing operation logs.

Using acquired data to train a model involves risks different from ordinary service provision. Distinguish whether personal information will be used for training or handled as statistical information or anonymously processed information. Simply changing the name is not enough; confirm the respective legal requirements and the explanations to be given to individuals and customer companies.

Distinguishing Entrustment, Third-Party Provision and Cross-Border Transfer

AI services generally use a combination of multiple external tools, such as external cloud services and machine learning APIs. In light of the APPI, carefully examine, based on how the external provider actually handles the data and the contract concluded, whether this constitutes "entrustment" or "third-party provision."

In particular, when adopting an AI API provided by an overseas provider, check whether it constitutes provision of personal data to a third party in a foreign country and, if so, whether the applicable consent of the individual has been obtained, the prescribed information has been provided, and contractual security control measures have been taken. Simply taking the form of entrustment does not automatically exempt you from the legal regulations that accompany cross-border transfers. You need to consider individually how much to disclose in the terms of use and privacy policy.

In a B2B model, cases where customer companies feed personal data they hold into your company's system can also be anticipated. Depending on the purpose of use and the actual handling, your company may be in the position of being entrusted by the customer with handling personal data. Also confirm the responsibilities you must bear as the entrusted party, such as the requirements for permission to sub-entrust and the performance of security control measures.

AI Output and Personal Information

In AI systems, examination is required not only of the data that is input but also of the output generated. Where AI evaluates individuals, classifies attributes, makes recommendations or summarizes information, that output may disadvantage the individual.

In areas directly connected to how individuals are treated, such as hiring decisions, personnel evaluations and credit screening, consider methods by which a person checks the AI's results and the reasons for the decision can be explained.

In addition, output estimated by AI is not necessarily true. If inaccurate personal information is recorded in a system as is or used in dealings with outside parties, it may lead to problems under the APPI or problems relating to creditworthiness and contracts.

Practical Operation After the Policy Is Established

Personal information protection for AI services does not end with publishing a privacy policy. Establish your operations to include control of access rights within the company, review of contracts with external vendors, and the initial response system in the event of an incident.

When your customers are large companies, you will be asked to answer security checklists and detailed questionnaires on personal information protection. If you are prepared to explain logically where data is stored, the scope of the purpose of use and the deletion procedures, it becomes easier to answer questions during business negotiations.

At LegalAgent, we see privacy legal work involving AI not as merely drafting rules, but as an issue bound up with service design and contract practice. We want legal to correctly grasp how data is actually used as technology advances, and to put in place explanations and management methods that match reality.