Skip to content

Glossary

APPIact on the protection of personal information

Japan's data-protection law, enforced by a national commission, which sets what a business must do when handling personal data belonging to people in Japan.

In plain terms

Japan's version of the rules most businesses now recognise: tell people what you are doing with their information, keep it safe, and do not quietly repurpose it. The shape will be familiar to anyone who has worked through a European exercise, and the detail is its own.

01

Why it matters

Because a business with staff or customers in Japan is handling material the law covers, and the AI tools now processing that material are usually operated by somebody else entirely. That turns a question about internal handling into a question about a third party, which is the point at which a routine product choice becomes a compliance one.

02

How it works

A single national commission administers and enforces it, which makes the regime easier to describe than one split across regional authorities. For a buyer the practical consequence is that guidance comes from one place rather than several, so the answers a vendor gives can be checked against a single source.

Handing material to a third party is treated as a distinct step rather than an incidental one. Using an external tool to process personal data is the kind of arrangement the law is interested in, which is why a vendor's terms rather than its features are the part that decides whether it can be used.

Moving material outside the country carries its own considerations, and most widely used AI tools process elsewhere. That makes the ordinary act of adopting a tool the thing the regime touches, rather than something that only matters at a later infrastructure stage.

It resembles other modern regimes closely enough to be misleading. A business that has done the work for one will recognise most of the structure here and should not assume the arrangements transfer, because the specifics of consent and third-party handling are where regimes diverge most.

For an organisation operating in several markets it is rarely the only regime in play. The practical question stops being which law to satisfy and becomes which vendor arrangement satisfies all of them, which narrows a shortlist faster than any individual requirement.

Where the regime touches a tool decision

Where the regime touches a tool decisionThe reason to separate these is that they are assessed by different people at different times, and the order is usually wrong. A team evaluates the left-hand column because that is what a trial shows, reaches a preference, and only then discovers that the preferred option cannot commit to what the right-hand column requires. Reversing the order costs almost nothing: the right-hand questions are three, they are answerable in writing, and a vendor that cannot answer them has told you something useful early rather than late. This is also the specific way the AI category differs from ordinary software procurement. In most categories the processing arrangement is an infrastructure detail that can be settled after the product decision, because the software runs where you put it. Here the material goes to the vendor by design, so where that happens and what may be done with it afterwards are properties of the product itself, and deferring them means deferring part of the product evaluation.What the product doesHow good the answers are.Which systems it connects to.What it costs per seat.What actually decidesWhat the vendor commits to inwriting.Where processing happens.What happens to materialafterwards.A shortlist built from theleft-hand column can failentirely on the right, and theright-hand column is invisibleon a pricing page. It is thepart that has to be asked for.
The reason to separate these is that they are assessed by different people at different times, and the order is usually wrong. A team evaluates the left-hand column because that is what a trial shows, reaches a preference, and only then discovers that the preferred option cannot commit to what the right-hand column requires. Reversing the order costs almost nothing: the right-hand questions are three, they are answerable in writing, and a vendor that cannot answer them has told you something useful early rather than late. This is also the specific way the AI category differs from ordinary software procurement. In most categories the processing arrangement is an infrastructure detail that can be settled after the product decision, because the software runs where you put it. Here the material goes to the vendor by design, so where that happens and what may be done with it afterwards are properties of the product itself, and deferring them means deferring part of the product evaluation.
03

Seen in the wild

  • Checking an assistant vendor's terms about third-party handling before a Japanese team adopts it.

    ChatGPT
  • Asking a search vendor where material is processed rather than where the company is registered.

    Glean
  • Keeping processing inside the country by running a model locally rather than sending material out.

    LM Studio
04

Common misconceptions

People assume

European compliance work covers it.

In fact

The regimes share a structure and diverge in the detail, particularly around consent and handing material to third parties. Work done for one is a useful starting position rather than an answer, and treating it as an answer is how gaps appear.

People assume

It only concerns companies based in Japan.

In fact

What matters is whose personal data is being handled rather than where the handling company sits, so an organisation with customers or staff there is worth checking regardless of whether it has a Japanese entity.

05

Questions

What does this change about choosing an AI tool?
It moves the decision from features to terms. Handing personal data to an external service is a distinct step the regime is interested in, so what the vendor commits to about processing, retention and onward transfer matters more than what the product does well.
Who actually enforces it?
A single national commission, which is a practical advantage when checking something. Guidance comes from one authority rather than a patchwork, so a vendor's characterisation of its own position can be compared against a single published source rather than several.
Does running a model ourselves avoid the question?
It removes the third-party step, which is a real simplification, and it does not remove the obligations. The organisation still handles the personal data and still owes the same duties about security and purpose; what changes is that nobody else is involved in the handling.
06

Key takeaways

  • Scope follows whose data is handled, not where the company sits.
  • Handing material to a third party is a distinct step, so vendor terms decide.
  • One national commission enforces it, which makes checking easier.
  • Familiar structure, different detail: other regimes do not transfer.
08

Tools that use this

  • ChatGPT

    Third-party handling terms before a Japanese team adopts it.

  • Glean

    Where material is processed, not where the company is registered.

  • LM Studio

    Running locally removes the third-party step entirely.

Last checked August 2026

All glossary terms