AI supports. People make the final call.
Miichisoft actively uses AI in development, while people remain responsible for quality, correctness, and confidentiality.
Why AI governance matters
AI governance is a set of risk management and operational rules for using AI safely and appropriately. It prepares for risks such as information leakage, hallucination (AI generating incorrect information), and bias (skew in training data), with people taking responsibility for quality, correctness, confidentiality, and legal and ethical considerations.
The more you use AI, the more questions arise: is confidential information protected, will AI errors go unchecked, are legal and accountability concerns covered. We address this through a structure where people make the final call at every stage, backed by preparation for these risks, so you can draw on the benefits of AI with confidence.
Risks we prepare for in using AI
- Information leakage and exposure of personal data. We treat all data as confidential under strict controls and run it in a Sandbox (an isolated execution environment).
- Misinformation from hallucination. People verify the basis for AI output at every stage and never adopt it as-is.
- Skewed judgment from bias. Important decisions are never left to AI; people make the final call.
- Accountability and legal compliance. We operate in line with your company policies and applicable law, keeping decisions and responsibility clear.
Four mechanisms for using AI safely
Quality, correctness, and confidentiality, protected by design.
People make the final call
AI produces the drafts, and we place a human confirmation gate at every stage. We enforce Human-in-the-loop (final judgment by a person) and never adopt AI output as-is.
We split each development stage across specialized AI agents, fixing procedures, notation, and output formats. This keeps deliverables at a consistent quality level regardless of individual staff experience.
Confidentiality in an isolated environment
We run AI in a securely isolated Sandbox (an isolated execution environment), operating on the premise that confidential information never leaves it.
No dependence on a single AI
We use multiple LLMs depending on the task, avoiding vendor lock-in (being tied to a specific product). Your development is not affected if one AI changes its specifications or is discontinued.
For details on access control, authentication, and data protection, see Security & Data Governance →
A human confirmation gate at every stage
AI produces the draft; people make the final call.
Three commitments on data handling
Information entrusted to us is handled strictly, within the scope of the contract and your permission.
Strictly managed as confidential
Entrusted information is kept under least-privilege access and controlled with audit logs.
Used only within the contractually defined scope
The scope of generative AI use is clearly agreed at the time of contract.
Requires per-project authorization
AI use is conditional on your request and approval. For projects with strict data protection requirements, we do not use AI.
No information leakage incidents have occurred on any project we have handled (our track record).
From PoC to production, and then in-house capability
We start with a PoC (small-scale proof of concept), grow it steadily, and leave the know-how with your company.
Start small
We limit the scope to verify effectiveness and safety.
Grow it steadily (production)
We expand into production while keeping the human confirmation gates in place.
Support bringing it in-house
On request, we leave the AI know-how with your company.
Three principles for responsible AI
The principles we hold to in order to use AI safely.
- 01
Transparency
We disclose to you which AI is used, at which stage, and how. Nothing operates as a black box.
- 02
Human-centered
Important decisions are made by people; AI remains purely a support tool. Ultimate accountability rests with people.
- 03
Compliance with laws and guidelines
We operate in line with your internal policies and applicable law, and reference AI guidelines from Japan and abroad.
Frequently asked questions
Will the data or code we provide be used to train AI models?
We manage all entrusted data as strictly confidential. The scope of generative AI use is clearly agreed in the contract, and each project requires your authorization (request and approval). For projects with strict requirements, we do not use AI.
Will we be locked into a specific AI service?
We use multiple LLMs depending on the task, avoiding vendor lock-in. Your development continues without interruption even if a specific AI changes its specifications or is discontinued.
Who holds the rights (IP) to the deliverables?
Ownership of intellectual property rights in deliverables is clearly defined in the contract. We propose a structure that lets you order with confidence.
Can we choose not to use AI at all?
Yes. For projects with strict data protection requirements, we operate without using AI.
If AI is involved, is quality still assured?
People make the final call at every stage (Human-in-the-loop). AI agents standardize output, and people pass each deliverable through the gates of requirements confirmation, design decisions, code review, and final verification. Our approach to assuring quality through numbers and structure is also detailed under Offshore Quality & Japanese-Language Support.
What about the risk of AI errors or bias?
To guard against hallucination and bias, important decisions are never left to AI; people make the final call. At every stage, people verify the basis for AI output and never adopt it as-is. We also limit the use cases and scope of impact.
How do you handle compliance with laws and internal policies?
We operate in line with your internal policies and applicable law. We reference AI guidelines from Japan and abroad, including the AI Business Operator Guidelines, and work to use AI responsibly.
Talk to us about how to approach AI adoption.
It is fine to start from "we want to use it but have concerns." We will propose a safe approach to AI adoption tailored to your requirements.
* The scope of generative AI use and data handling is based on the contract and your authorization (request and approval).
* Ownership of intellectual property rights in deliverables is defined individually in the contract.
* "Zero information leakage incidents" is a cumulative figure and our own track record.
* Standardization through AI agents, isolated execution in a Sandbox, and the use of multiple LLMs are standard practices based on our development process.
* We reference guidelines such as Japan’s AI Business Operator Guidelines (issued by the Ministry of Internal Affairs and Communications and the Ministry of Economy, Trade and Industry) and work to use AI responsibly. This does not indicate acquisition of any specific certification.
* The contracting entity is our Japanese corporation (Miichisoft Japan Co., Ltd.). Bilingual support (JA/EN) is available.