Here at LOTI, we have worked on a guide to procurement, implementation and ongoing use of AI-based technologies, which provides practical steps and recommendations for best practice in compliance with data protection law and the Public Sector Equality Duty when taking steps to procure or commission AI-based technologies. But how can AI be used to improve the digitalisation of legislation processes? Polly Kwok, Communications and Engagement Manager at LOTI sits down with the Legal Team at Japan Digital Agency to learn more about how they’re approaching legislation digitalisation challenges and using AI to help. 

Polly: For our LOTI audience, who may not know, what is the Japan Digital Agency?

JDA: The Japan Digital Agency (JDA) launched in September 2021, established to accelerate digital transformation across the Japanese government and society, serving as a digitalisation command tower for each ministry, coordinating planning, standards, and government IT investment across ministries. Their mission is to achieve “Human-friendly digitalisation: No one left behind”. Within the Legal team, our main areas of work include providing user-friendly legal information and promoting new business and services using legal data. 

Polly: With a traditionally heavily paper-based system around legislation, what are some initiatives you are doing within the Legal team to promote digitalisation and sharing of new ideas and practices?

JDA: Understanding of legislation in Japan is highly specialised, and there are not many people who are familiar with both the law and the latest digital developments. So we have started new initiatives around bringing different professionals together to tackle our digitalisation challenges. For example, since 2023 we have organised hackathons using legal data and APIs, including the “Law API Hackathon” and the “Law × Digital Hackathon.” In the 2025 “Law × Digital Hackathon”, 28 teams participated and submitted their prototype tools. Attendees included engineers from the private sector, lawyers, students and government, creating 28 tools. One of these tools was an AI tool that reads the building rules, and then automatically creates the checking steps, and then verifies whether a building design follows those rules. We have also created a new Legal Data Community, bringing together field practitioners with real issues and participating engineers exploring solutions using open regulatory data.

Polly: In local government, there are challenges around “Analog Debt”—outdated local bylaws that hinder digital-first services, for example requiring individuals to check for flooding instead of cameras/sensors. How did JDA approach this issue as part of the digitalisation process?

JDA: That is a great question and that is something we are also working on. The first step was to identify what these laws are. 

In Japan, the government conducted a comprehensive review of existing legislation and identified thousands of provisions that imposed analogue requirements such as paper documents, on-site inspections, or face-to-face procedures. A vast majority of these regulations have since been revised or removed as part of the digitalisation process.

In addition, we have introduced a digital legislative review function—sometimes referred to internally as a “Digital Legislation Bureau”—which checks all new bills submitted to the government to ensure they comply with the government’s Digital Principles and do not introduce new analogue requirements.

To support this process, we recently developed a generative-AI-based tool that helps screen draft legislation for potential analogue requirements, and this tool has begun to be used internally in the review process.

Polly: You’ve mentioned that you’ve developed a generative AI system to review “Analog regulations”, how have you managed the legal risks of using generative AI for statutory review? For example, even a punctuation change to a comma in law can completely change the meaning. 

JDA: This is an issue our Legal team has been focusing on recently. When using AI for tasks related to legislation, a high level of accuracy is required. However, by its nature, generative AI can never be expected to produce perfectly accurate results all the time. In addition, when dealing with highly specialised legal texts, we feel that the accuracy of general-purpose AI systems is still limited.

For this reason, the key is to design workflows that incorporate human-in-the-loop verification. In our review of analogue regulations, we have established a process in which the results generated by AI are always subject to final human review. This allows us to use generative AI in practice while managing the risk that some provisions may be overlooked, recognising that AI outputs will not always be perfect.

In our view, the most important point is to recognise that generative AI can make mistakes, and to design workflows that remain reliable even under that assumption.

Polly: So what is next for the JDA Legal team digitalisation journey of legal data?

JDA: We are looking into the use of AI-assisted legislative drafting by combining retrieval-augmented generation (RAG) with structured legal databases. In Japanese administration, Legislative drafting is widely recognised as a complex and time‑consuming process, requiring careful use of established legal phrasing, consistency across related laws, and respect for long‑standing drafting conventions—expertise that is usually built up through experience. By using AI to help identify relevant legal provisions and suggest draft text based on reliable sources, we hope to ease some of the cognitive and practical burden on drafters.

Polly: We look forward to hearing more about what the JDA Legal team does next! What are some key takeaways you would share from using AI in the legal context with our audience?

JDA: For us, we want to stop the creation of new “Analog regulations” in the first place, so that we can start with a “Digital First” approach. From a digital transformation perspective, it is important not to build in the process now that will hinder digitalisation moving forward. Secondly, we have found that AI works best for the legal use cases for triage, and not decision making. We are using LLMs (Large Language Models) to scan massive policy documents for specific friction points, keeping legal experts in the loop so that the decision-making is human-driven. Lastly, we know that legal professionals are not AI experts, so it is about bringing together different stakeholders who may have the tech expertise to help solve and look at our issues from a different perspective. 

 This blog was co-written with the following members of the Japan Digital Agency Legal Team: Tomoki Ueda (Legal Specialist), Yuki Odani (Official) and Daiki Kogure (Official).  To find out more about the work of the Japan Digital Agency, please visit their website.