Using AI for Contract Review and Article 72 of the Attorney Act: What Legal Teams Can Leave to AI
Hello, I'm Noriaki Asato, Representative Attorney at LegalAgent.
Late in the afternoon, a business unit sends over a contract for a new deal and asks for an urgent review. With the deadline approaching, the in-house legal staffer loads the full text into a generative AI tool, and within moments the screen shows proposed clause revisions and a list of issues of concern. The question that arises at this point is whether those comments can be trusted as they are and sent back to the counterparty. Simply looking at the text on the screen does not tell you whether the output reflects the particular features of your business or your past dealings with the counterparty, or whether it accounts for the balance of bargaining power. The sense that the work has become faster and the uncertainty about how to turn this output into your company's own judgment arise at the same time.
The other source of uncertainty lies on the legal side. Article 72 of the Attorney Act prohibits anyone who is not an attorney from handling legal services as a business for compensation. How does this provision relate to having AI review contracts, or to providing AI contract review services to companies? The Ministry of Justice published guidelines on this point in 2023 and in 2026, and reading these two documents gives the broad outline of how far AI can be used in legal work.
Where AI Is Being Used in Legal Work
The situations in which AI is used in corporate legal work can be divided as follows, according to the role AI plays and the stage of work it covers.
- Software that supports the drafting, review and management of contracts
- Research and summarization of statutes, case law and internal materials
- Support for drafting documents such as minutes, internal explanatory memos and translations
- Entering contracts or questions into general-purpose conversational generative AI
- AI agents that autonomously carry out multiple work steps according to instructions
The guidelines published by the Ministry of Justice in 2026 refer to services that use these technologies as "AI and other legal work support services." They are not limited to dedicated software specialized in contract review; the guidelines also have in mind cases in which a provider of general-purpose AI offers its service to businesses for consideration.
What AI does well is high-volume preparatory work that follows a set format. This includes extracting issues by identifying unfavorable or missing clauses across an entire contract, comparing clauses against your own template or the counterparty's proposal, cross-checking redlines, extracting and summarizing relevant passages from case law and administrative materials, and preparing first drafts of proposed clause revisions.
On the other hand, judgments that depend on information outside the contract are made by people. The capital relationship and balance of power with the counterparty, the track record of past transactions, the business imperative to close this particular deal, the level of risk the company can tolerate, and the depth of explanation that will satisfy management are not written anywhere in the contract language. Whether to accept the risks AI has identified in light of the company's own standards, and the negotiating strategy of which revisions to propose to the counterparty and which demands to withdraw, are matters decided by the legal staff or attorneys who have this information. The term "AI lawyer" is sometimes used, but AI is not an entity that assumes legal responsibility for judgments. The meaning of the term and how to divide roles between attorneys and AI are covered in What Is an "AI Lawyer"? How to Divide Roles Between Attorneys and AI in Corporate Legal Work in the Generative AI Era.
Article 72 of the Attorney Act and AI Contract Review
The unavoidable legal question in using AI for legal work is its relationship with Article 72 of the Attorney Act. The main text of that Article provides as follows.
"No person other than an attorney or a legal professional corporation may, for the purpose of obtaining compensation, engage in the business of providing legal opinions, representation, arbitration, settlement or other legal services in connection with litigation cases, non-contentious cases, cases of objection against administrative agencies such as requests for review, requests for re-investigation and requests for re-examination, or other general legal cases, or acting as an intermediary for such services."
The proviso to that Article states, "This shall not apply where otherwise provided in this Act or other laws."
Whether an AI service provided by a private business constitutes the unauthorized practice of law prohibited by this provision is a question of whether the elements of a penal provision are satisfied. It is judged on the individual facts in light of the purpose of the Article, namely that allowing persons other than attorneys to make a business of intervening in other people's legal cases would harm the interests of the parties and other persons concerned, obstruct the fair and smooth conduct of legal life and, in turn, harm the legal order (Supreme Court, Grand Bench judgment of July 14, 1971), and it is ultimately left to the courts.
The Ministry of Justice has published guidelines on this interpretation twice: "On the Relationship Between the Provision of Contract-Related Work Support Services Using AI, etc. and Article 72 of the Attorney Act" in August 2023 (the "2023 Guidelines"), and "On the Relationship Between the Provision of AI and Other Legal Work Support Services in the Business Field and Article 72 of the Attorney Act" on August 21, 2026 (the "2026 Guidelines"). The 2026 Guidelines supplement and expand the 2023 Guidelines in response to advances in generative AI and the spread of conversational and autonomous services, and the two are read together.
The 2023 Guidelines set out the general approach for each element of Article 72.
"Compensation" in "for the purpose of obtaining compensation" means consideration for the service of handling legal matters. Where a service is provided without receiving any benefit at all, such as a usage fee, this element is usually not met. On the other hand, it is understood that the element may be met where there is in substance a relationship of consideration, such as steering users to a paid service, payment from a third party, or providing the service only to those who have paid membership fees or the like.
"Other general legal cases" means matters in which there is a dispute or doubt regarding legal rights and obligations, or in which a new relationship of rights and obligations arises, and it is considered that "case-ness" is required, meaning a degree of dispute or doubt comparable to a litigation case. Case-ness may be found in preparing a settlement agreement after a dispute has arisen between the parties to a transaction. A contract clarifying customary transactions between a parent and subsidiary, or a procurement contract under a master agreement on the same terms as before, is usually considered difficult to regard as having case-ness. Discussions toward concluding a contract and the review of legal issues that accompany the ordinary work of corporate legal departments are judged by considering the purpose of the contract, the relationship between the parties and the background, while bearing in mind the authorities' observation that in many cases these lack case-ness.
In "legal opinions... or other legal services," a legal opinion means stating a legal view based on specialized legal knowledge, and other legal services means handling matters that create, modify or otherwise affect legal effects. Whether a service falls under this is determined from its functions and the way it is presented to users. The function-by-function comparison that the 2023 Guidelines set out for contract review services is as follows.
| Type of service | Functions that may fall under it | Functions that usually do not fall under it |
|---|---|---|
| Contract drafting support | Displaying a specific contract by legally processing the background and circumstances leading to the contract in response to a user's non-standard input | Merely selecting a registered template according to standard input or choices and displaying it as is, or with the input reflected |
| Contract review support | Displaying the existence or degree of legal risk according to the individual matter; displaying specific proposed revisions that legally process the background and circumstances | Displaying differences from a registered template without regard to the meaning of the words; displaying similar parts based only on linguistic similarity; displaying general sample clauses or explanations linked to a template or checklist without individual modification |
| Contract management support | Automatically displaying, from time to time, the legal risks or need for action according to the individual matter for managed contracts | Classifying and displaying according to terms such as the contracting parties, contract date, performance date, renewal date and amount; displaying registered points of attention when a registered time or condition is reached |
The guidelines also address cases where the user is an attorney. Where the service is provided to an attorney or legal professional corporation and the attorney uses it by personally scrutinizing the results, taking them into account, and making revisions as necessary, and likewise where the service is provided to a non-attorney but an attorney who is an employee or officer of the recipient uses it in the same way, it is understood that the use usually does not violate Article 72.
Reflecting the realities of conversational AI and general-purpose models, the 2026 Guidelines supplemented the analysis of whether a service "handles legal opinions... or other legal services" in connection with matters that have case-ness. Even where a user enters a prompt and has the AI generate output, if the service provider substantively leads or intends that outcome through the design and provision of the service, the act may, on a normative assessment, be regarded as the service provider's act. Case-ness is judged, with reference to the Supreme Court decision of July 20, 2010, by the standard of whether the matter is one in which a legal dispute can be expected to arise almost unavoidably.
On the basis of this standard, the 2026 Guidelines introduced the concept of "value-neutral service provision." A service may be called value-neutral where its design and functions are not aimed at having it used for legal services relating to matters in which a legal dispute has already surfaced or is almost unavoidable, and it does not have characteristics or functions specialized for or intended for such use (such as screens whose main purpose is handling matters with case-ness, or displays that actively promote that use). A chatbot function that allows users to upload files and freely enter prompts falls within this scope.
Even for a value-neutral service, the risk of liability for violating Article 72 cannot be ruled out where the provider continues to provide the service while knowing and accepting that it is actually being used for matters with case-ness, or while knowing and accepting a situation in which there is a high probability that a not-merely-exceptional range of users will use it for such matters, without taking reasonable measures such as suspending use, issuing warnings or directing users to consult an attorney.
Services intended for support that accompanies ordinary business, specifically research of literature and other sources and review of legal issues; drafting, review and management of documents; governance, risk management and compliance, including internal training and confirming the legality of management decisions; investigation of whistleblowing matters; building new business schemes and business reorganizations; internal investigations to decide the company's response policy when trouble arises; and meeting work such as shareholders' meetings and board of directors meetings, were classified as generally capable of being called value-neutral service provision, as long as their design and functions do not fall under the above.
A line is also drawn regarding the identification of legal risks. A general observation such as "this clause may give rise to litigation risk" is not in itself considered to immediately constitute legal advice on a matter with case-ness, but it may do so where it is made in connection with a specific dispute. Drafting documents for external use that presuppose a legal dispute, such as complaints, briefs, answers, statements of evidence and other court filings, or settlement agreements, falls under it.
The 2026 Guidelines also list governance-related points of attention and recommendations for service providers: establishing an incident response contact point in Japan; substantive involvement of Japanese attorneys in designing functions that may amount to legal opinions; explaining the nature of retrieval-augmented generation (RAG), training data and output; not using expressions that suggest the service provides judgments in place of an attorney or guarantees the accuracy of legal conclusions; confirming users' attributes and purposes at the time of application; stating matters expressly in the terms of use; displaying disclaimers before prompt entry and on output; strengthening cautionary displays for high-risk functions; displaying grounds and sources; directing users to the Japan Federation of Bar Associations' attorney search; and suspending use in cases of serious breach of the terms. The matters to be stated expressly in the terms of use include that AI does not make final judgments in place of attorneys or other professionals and its answers are only reference information; that the service is not provided for the purpose of being used for matters with case-ness; that such matters need to be taken to an attorney; and that the service is to be used only for the user's own business and not for handling other people's legal services.
The use of a service by a parent company's legal department to support the legal work of subsidiaries and affiliates within a corporate group was classified as usually not falling under "handling other people's legal services." Cases in which an attorney or legal professional corporation develops and provides a service itself, or provides legal services by making judgments in the attorney's name and on the attorney's responsibility based on the service's output (the attorney-provision model), are considered not to immediately conflict with Article 72. Note that the 2026 Guidelines cover the "business field," meaning use by businesses in connection with their own business activities, and services mainly aimed at consumers may require separate consideration.
How Corporate Legal Departments and Law Firms Each Use AI
Applying the framework in the previous section to corporate legal departments gives the following picture. When a company's legal department or legal staff use AI to review contracts to which the company is a party or for the company's own work, they are handling their own business, not other people's legal services. In organizations with in-house attorneys, the attorney-use model can be adopted, in which the attorney personally scrutinizes and revises the results while taking them into account. In organizations without in-house attorneys, the practice should be to follow the terms of use and disclaimers, not to process matters with case-ness through AI, such as matters in which a dispute has surfaced, matters in which rights and obligations are being contested with the counterparty, or matters requiring a settlement agreement, and to consult an attorney instead.
When a law firm such as LegalAgent uses generative AI or AI agents for contract review and legal consultations, the attorney takes the AI output as a starting point but personally confirms the facts and makes the legal judgment, and gives the client an opinion in the attorney's own name and on the attorney's own responsibility. Under this attorney-provision model, matters with case-ness can also be handled.
The steps for a company to embed AI into contract review and research should be designed in sequence.
The first step is to separate uses. Start with ordinary work without case-ness, specifically the first read of contract reviews, clause comparisons, research of statutes and case law, drafting minutes and preparing internal explanatory memos. Exclude dispute response, drafting settlement agreements and preparing litigation documents, and decide in advance the criteria for passing a matter to an attorney once a conflict with the counterparty arises.
Next, check information management. Confirm, both technically and contractually, that the settings prevent input data from being used for training, and scrutinize whether information subject to confidentiality obligations under a non-disclosure agreement may be entered. When handling documents containing personal information, check whether this constitutes entrusting personal data. Gaps that tend to arise in recording meetings and entering data are covered in AI Meeting Minutes Tools and Confidentiality Obligations, and how to create internal rules is covered in How to Create a Generative AI Use Policy.
When vetting vendors, the points of attention and recommendations in the 2026 Guidelines can be used directly as question items. Confirm whether there is a contact point in Japan, whether attorneys are involved in the design, whether grounds and sources are displayed, whether the disclaimer and terms of use state that the service is not to be used for matters with case-ness, and whether there are rules on suspending use. How to proceed with contract vetting is covered in AI Vendor Due Diligence and Contract Review.
Also decide the procedure for checking output. Article numbers and sources are checked by a person against primary materials. Prioritizing the issues AI raises and deciding whether to accept them in light of the company's risk tolerance is the job of the legal staff or attorneys. Proposed revisions and comments for the counterparty are finished by a person who adjusts them to the context. The overall flow of contract review is summarized in What Is Contract Review? How to Proceed with Contract Review in the AI Era.
For internal rules and records, start by writing into the generative AI use policy the work for which AI may be used, the information that may be entered, who checks the output and where matters are escalated. A system for recording the reasons for judgments and carrying them over to the next matter is covered in Legal Playbook.
When using AI agents, decide in advance how much execution authority, such as placing orders or sending messages, to grant them, and who checks and stops them and when. Authority design is covered in Designing Authority for AI Agents, and an entry point to using developer-oriented AI agents for legal work is covered in Codex and Claude Code in Legal Work.
As a task you can do the day after receiving a draft, one approach is to classify your company's legal work into three categories, "routine work without case-ness," "work requiring judgment" and "dispute response," and create a list that limits the scope entrusted to AI to the first category. If you separate at the outset the areas you entrust to AI and the areas in which people make judgments, the foundation of your operations will not change even when you replace tools.
AI in Legal Work and the Involvement of a Law Firm
At LegalAgent, our attorneys use AI agents to conduct contract reviews and legal consultations. We support both the design of AI use in corporate legal departments, specifically separating uses, drafting rules and vetting vendors, and matters that require an attorney's judgment.
Frequently asked questions
Does having AI review contracts violate Article 72 of the Attorney Act?
When a company, as part of its own business, uses an AI tool to review contracts to which it is a party, it is not handling other people's legal services, so this does not immediately violate Article 72 of the Attorney Act. On the vendor side that provides the service, however, where the service judges legal risks or presents proposed revisions tailored to individual matters, there is a risk of conflicting with that Article depending on whether there is compensation and whether there is case-ness. User companies are also expected to operate on the basis that matters involving disputes are taken to an attorney.
What is "case-ness" under Article 72 of the Attorney Act?
It refers to the nature of a matter in which there is a dispute or doubt regarding rights and obligations, or in which a legal dispute can be expected to arise almost unavoidably. The Ministry of Justice guidelines of 2023 and 2026 treat as a requirement the existence of a dispute or doubt to a degree comparable to a litigation case. Preparing a settlement agreement in a matter where a dispute has already arisen is understood to involve case-ness, but negotiations toward concluding contracts that accompany the ordinary work of corporate legal departments, and routine transactional contracts, are considered in many cases to lack case-ness.
Are there legal tasks that should not be left to AI?
Handling matters in which a dispute has arisen, deciding the negotiation strategy with the counterparty, and making final judgments in light of the company's risk tolerance are decided by legal staff or attorneys rather than left to AI. Drafting court filings such as complaints and answers, and drafting settlement agreements, are tasks that the Ministry of Justice guidelines indicate may constitute legal services relating to matters with case-ness. Judgments on whether terms are acceptable, made in light of circumstances outside the text of the contract such as the background of the transaction, the balance of power between the parties and the impact on management, are considered something AI cannot replace.
What should be checked when choosing an AI contract review service?
A useful benchmark is whether the service complies with the governance standards set out in the guidelines. Specifically, these include whether there is a contact point in Japan, whether attorneys are involved in the design, whether there is a function displaying the grounds for the output, and whether input data can be set not to be used for machine learning. In addition, checking whether the terms of use and disclaimers expressly state that the service is not to be used for matters with case-ness, and whether rules on suspending use in the event of a breach of the terms are in place, forms the basis for safe operation.
Services connected to this topic