Rakuten Securities disclosed on November 17, 2026 that an attacker reached one of its test environments on November 11 and may have taken data on millions of accounts, including names, addresses, phone numbers, employers, withdrawal bank details, and self-reported income and assets. Passwords and trade PINs were not compromised, and the company has confirmed no unauthorized transactions. The detail that should worry every business in Japan is not the attacker. It is that unmasked customer data was sitting in a vulnerable test environment because the company put it there, through a formal, approved internal procedure.

What happened in the Rakuten Securities breach?

On Monday, November 16, Rakuten Securities detected signs of unauthorized access to part of its systems and cut off access to the affected area. Its investigation traced the intrusion to November 11, and to a test environment. The company disclosed the breach publicly on Tuesday, November 17. The Japan Times describes it as the largest hack of a Japanese securities company to date.

The potential exposure comes in two categories. Up to about 11.25 million accounts: branch codes, customer codes, and login IDs. Up to about 4.47 million accounts, including former customers: name, date of birth, address, phone number, email address, employer, withdrawal bank name, account number and account holder, self-reported annual income and financial assets, holdings by product, and last login date. The company lists these as separate categories, and how much they overlap is unstated, so they should not be added together into one total.

The company says this information may have been taken; confirmed theft of every record is not established. The cause and the exact access route are still under investigation. Anyone claiming to know precisely how the attacker got in is ahead of the evidence. What is not under investigation is where the data was. No, where the data was is blatantly obvious.

Why was customer data in a test environment?

ITmedia asked the company directly, and the answers are the most revealing part of this entire incident. During a system upgrade, some functions needed to be verified against production data. So a portion of real customer data was copied onto a test server for a period of time. It was not masked.

The company stressed that this was not one employee acting alone: it went through a formal internal procedure. Normally, it says, test environments hold only dummy data, and this was an exceptional case. It declined to say when the data was moved or how long it sat there. Asked whether the same thing had happened before, it answered: almost none, this was a special case. It said security measures were in place, but that a vulnerability was exploited.

Hiroaki Kuramochi, CTO of the security company LAC, told ITmedia that managing customer information in a test environment is not standard practice. The principle is dummy data, and if production data must be reused, proper masking is the rule. He was charitable about the motive: building test cases that cover every kind of real customer registration is hard, and real data is tempting. I suppose the temptation is understandable, but only if you’re that lazy or incompetent. Regardless, don’t give in to it. Or, if you do, understand the extremely damning implications of your failure.

The governance failure should be the headline

Most of the coverage of this breach will chase the attacker. Fine. The investigation hasn’t established how the attacker got in. What is already clear is that Rakuten Securities had copied real customer data into a test environment without masking it.

Across the industry, test environments are typically run to different standards than production. They change constantly, more people can touch them, and they are built to be broken, because that is their job. That is exactly why the dummy-data principle exists. What standards Rakuten Securities’ own test environment met, the company has not said.

So the questions we all should have are procedural, and they were all answered before any attacker appeared. What justified using live customer data instead of masked or dummy data? Who signed the exception? What controls limited who could reach the copy? What guaranteed it was deleted afterward? “It went through formal internal procedure” is not a defense. It is the diagnosis of a weak and scared operational mandate. A procedure looked at unmasked production data in a test environment, and said “yes”.

A company can do everything right on the production side and still lose the same data through a copy. Whether anyone owned this one, and what protections were approved alongside it, are questions the investigation should answer in public.

What did the attackers actually get?

Not passwords, and not the PINs used to execute trades. The company is clear that those were not compromised, and, ok, great. But look at what the 4.47 million account category may include: your name, your date of birth, your address, your phone number, your employer, your bank, your account number, your income, and your holdings.

With that, an attacker does not need to break into your brokerage account. They can call you, sound exactly like your broker, quote your own account details back to you, and walk you into moving the money yourself. Rakuten’s own disclosure warns customers about exactly this: emails, SMS, phone calls, and even physical mail impersonating the company. That warning is the company acknowledging the real shape of the risk.

This, by the way, is known as phishing.

“No unauthorized transactions confirmed” deserves to be stated exactly as what it is: the company has not found unauthorized trades as of disclosure. That is reassuring, and it is also the narrowest possible measure of harm. It says nothing about what the exposed data enables over the next year, in patient hands. It also downplays the severity of the operational failure. A company that does things like this deserves to have its Trust Points severely reduced.

And here’s some salt on the wound: to find out whether your information may have been exposed, you have to log in and check a notice screen. Former customers cannot log in; the company says it is working on notifying them by some means, without details. If you closed your account years ago, your data was still in the building, and the clarity of your answer now depends on a notification method the company has not described.

Are passkeys enough to protect customers?

Credit where it is due: more than 90 percent of Rakuten Securities’ daily-login users already authenticate with FIDO2 passkeys, and the company says it is moving toward making them mandatory. That is genuinely good, and the right direction. Passkeys kill off whole categories of phishing.

But a passkey protects the login to the real system. It does nothing for a copy of the data sitting on a test server. Let me reiterate that the company itself duplicated real customer data and relocated it to a vulnerable environment outside of the control of regular channels. Locking the front door does not help if the filing cabinet was carried out the side door during renovations. These are two different problems, and only one of them is solved by better authentication.

Is this only a Rakuten problem?

No, and singling out one company would miss the point. This is the third Rakuten Group disclosure in weeks: Rakuten Travel reported two incidents covering more than 2.6 million accounts, and Rakuten Drive disclosed that data from over 15,000 accounts was exposed in a window running from January 29 to September 17. SMBC Nikko Securities reported over 56,000 accounts compromised last week, and Daiwa Securities, IwaiCosmo Securities, and Sony Bank have all dealt with leaks through external service providers. Rakuten Group shares dropped more than 7 percent on Wednesday morning before recovering.

I have written before about breaches where the damage traced back to ordinary operating decisions: card numbers kept after their purpose ended, customer data handed to providers whose systems then served someone else, and keys stored where a public status dashboard could betray them. Rakuten Securities adds a sharper case: an objectively wrong operating decision that was formally approved. The procedure worked as designed. The design was the problem.

How big is Japan’s breach problem right now?

Bigger than any single headline. I maintain the Japan Cybersecurity Tracker, which tracks publicly disclosed cyber incidents in Japan, one row each, with sources. As of November 17 it held 1,708 incidents disclosed since June: 211 of them in the seven days from November 11 to 17 alone, 64.1 percent of them coming in through a vendor, and 121 compromised vendors named. In the same week Rakuten Securities was working out what it had lost, vendor breaches at KARTE, OpenLogi, and COMSBI were rippling through dozens of other companies at once.

Rakuten Securities is not an outlier. It is a well-documented example of a pattern its rows repeat week after week: copies, vendors, and exceptions. Browse the tracker and judge the scale for yourself.

Where is your customer data copied right now?

That is the question that this incident now makes even more imperative, on top of the more than 1,700 cyber incidents publicly disclosed in Japan since June. How quickly could you answer it?

Start with an inventory. Where do copies of your customer data exist outside production: test servers, analytics exports, vendor environments, a spreadsheet on someone’s desktop. For each copy, who approved it, and is the reason still valid?

Make masked or dummy data the default. If real data is genuinely necessary, the exception should come with an expiry date, a named owner, and the same protection as production. If nobody can tell you when a copy gets deleted, that copy should not exist.

Then make the environment itself boring. Test systems should be reachable by the fewest people possible, logged like production, and reviewed on a schedule. A copy made for one verification gets its deletion date before it is created, and deletion is confirmed, not assumed. Vendors get the same questions in writing: where does our data go, who can touch it, and when is it destroyed. If the answers are vague, that is your answer.

Then ask your vendors the same questions, because as Daiwa and Sony Bank just demonstrated, your data’s worst day may happen in someone else’s building.

A test environment does not make the customer data inside it any less real.

Own your data and wrangle your operations

If this piece raised a question about where your own customer data lives, that question is worth an hour. A Decision Session is 45 minutes, live, on your systems and the decision in front of you. If you want the picture in writing, Diagnostics maps the tools, ownership, and workflows you already run, and hands you a report and a debrief. Both are described on my services page, and you can contact me directly. Tell me what you need to decide.

And if you want the evidence behind the pattern first, every publicly disclosed incident is on the Japan Cybersecurity Tracker, one row each, with sources.