Chaen.ai Work with me

Japan market

Muji Data Breach: Card Numbers Used as Customer IDs

🗓 ⏱ 5 min read

中文日本語

Muji Data Breach: Card Numbers Used as Customer IDs

Muji reported a data breach affecting up to 27.26 million customer records. Full credit card numbers were used as customer IDs until 2021, stored unencrypted.…

Muji has reported that up to roughly 27.26 million customer records may have been exposed to unauthorized access, and that about 1.09 million of those also included credit card numbers. The part that stopped me was not the attack. It was the design: until 2021, full credit card numbers were used directly as customer identification IDs, stored with no encryption and no masking. This piece sorts out what has actually been disclosed and then turns it into checks any company can run on its own systems this week.

What the Muji data breach has actually disclosed

These are the only points stated as fact:

  • Up to roughly 27.26 million customer records may have been accessed without authorization
  • About 1.09 million of those also include credit card numbers
  • Until 2021, the full credit card number was used as the customer identification ID
  • Those card numbers were stored without encryption and without masking
  • After the system was changed they were not deleted and remained for roughly five years

The intrusion path, the technique, and the response are not known from what has been reported, so I am not going to guess at them. Filling gaps with speculation is how a useful discussion turns into noise. What remains is two issues: how the identifier was designed, and the fact that data nobody used anymore was never deleted. Both are upstream of any attacker. Both can be fixed inside your own company without knowing anything about who attacked whom. Separating the before-the-attack problems from the after-the-attack problems is the fastest way to make an internal security discussion productive.

Why a credit card number is the worst possible customer ID

An identifier is the single most widely copied value in any system. Put it in a primary key and it shows up in logs. It rides in URL parameters. It syncs to the analytics warehouse. It goes out as a CSV to the sales team. It is passed to connected services. It gets written to intermediate files by the nightly batch. It essentially never stays only in the places the original designer imagined.

So the moment you choose a value as an identifier, you have given up the defense of limiting where it is stored. Encryption and masking are about putting a lock on the vault. An identifier is a value that is designed to walk around outside the vault. You can encrypt the main table as carefully as you like, and the copies you handed out as IDs will still leak first. For me this is the heaviest point in the whole story.

On top of that, a card number has terrible properties as an identifier in the first place. The customer cannot freely change it, it changes anyway when the card is reissued, and the same number is in use at other companies’ services. What you want from an identifier is a meaningless value you can mint freely in-house and throw away when needed. A card number is the exact opposite. A sequence number or random string generated internally means nothing on its own even if it leaks. That difference cannot be patched in later with a feature.

So the first question is not whether the data was encrypted. It is what you chose as the ID.

The five-year-old data nobody deleted

The second issue is that after the practice changed in 2021, the actual data was not deleted and stayed for roughly five years. I don’t think this is unusual at all. Migration plans are written carefully up to the point where the new system runs. What happens to the old data, who deletes it, and when, tends to stay vague. Once the new system is live the project is declared a success and the old tables belong to nobody.

The mental model that matters here is that “not used” and “not present” are completely different states. A table nobody references is invisible in day-to-day operations. It doesn’t show up in monitoring. No code calls it, so grepping the codebase won’t surface it. And it still lands in the backup every single night. Until it is actually deleted, it remains fully in scope for a breach.

The fix is unglamorous: make disposal of old data an explicit step in the migration plan, and keep a record that the disposal actually ran. “We were going to delete it” is not deletion. A log of who deleted what and when is what lets you say it is gone.

A checklist companies can run this week

In rough priority order:

  • Are you using sensitive values in primary or foreign keys? Card numbers are the obvious case, but phone numbers and email addresses fall into the same pattern
  • Has that identifier been copied into logs, the analytics warehouse, or BI tools? Checking only the main table is not enough
  • Do the old tables and dump files left over from past migrations have a defined retention period?
  • Is there a record that deletion was actually executed? Is there an inventory to check against?
  • Can you produce a list of which identifiers you hand to outsourcing partners and SaaS vendors?

The first two can usually be answered the same day. Read the schema definitions, search the log output. If nothing turns up, you can at least say you are not in the same pattern as this incident. Items three onward are the heavy ones, and the older the system — especially when whoever built it has long since left — the longer it takes. Which is exactly why it is worth starting before the next new initiative.

How this lands in Japan and what overseas vendors should take from it

In Japan, scrutiny tends to fall less on how sophisticated the attack was and more on whether the basics were in place. That held here: the reaction centered not on attack technique but on the single fact of storing full card numbers. The fact that it is a familiar retail name matters too. When it is a store people use in daily life, the story is received as a story about their own card.

That structure carries directly into how overseas AI and security companies should talk to the Japanese market. Japanese enterprises often run core systems that have been live for a very long time, frequently with nobody left in-house who built them. What lands is not the newest detection feature on its own, but the translation of that feature into “here is how you find the old design decisions buried in your systems.” Telling a Japanese decision-maker that they can surface where sensitive values are embedded as identifiers is far more persuasive than a generic security pitch. The same applies to disclosure style: stating concretely what was stored earns more trust than leading with an apology.

How I would apply this inside my own company

The first thing I would do is an inventory of the schema: list what value is being used to uniquely identify a customer. Then trace how far that value travels — into logs, into analytics, into external integrations. Evaluating a new security product can come after that. If you don’t know what is where, you cannot decide what to protect. Rather than consuming this as someone else’s failure, I want to use it as the trigger to actually run the identifier-design review and the old-data inventory we keep postponing.

FAQ

How many records were affected in the Muji data breach?

According to what has been disclosed, up to roughly 27.26 million customer records may have been accessed. Of those, about 1.09 million also included credit card numbers.

Why were credit card numbers used as customer IDs?

The reason has not been explained in what is currently public. What is known is that the practice continued until 2021 and the numbers were stored without encryption or masking.

How long did the old card data remain in the system?

It was not deleted after the system was changed and remained for roughly five years. Nothing beyond that about the timeline is known from what has been reported.

Based on a post on XOriginal post ↗

Share this article

XFacebookLINEはてブ

About the author

Masahiro Chaen
Masahiro Chaen Founder & CEO, DigiRise Inc.

CEO of DigiRise Inc. Has supported 600+ companies and 50,000+ people with AI training, consulting and development. GMO AI & Crypto advisor, Google Gemini AI advisor. 200K+ followers on X.