Japan’s September 2026 breaches: a briefing

Matt Ketchum, MKUltraman · Scope: incidents disclosed 1–30 September 2026 · Data as of 16 November 2026 · Source: the Japan Cybersecurity Tracker

Written in November 2026 from tracker data and dated to the period it covers. It was not published at the time.

The 30-second briefing

Key facts. Organisations in Japan published 99 breach notices in September 2026, up 5 on August’s 94. They trace back to about 91 distinct breaches, the closest the two numbers came all summer. Only 23 of 99 (23.2%) came through a supplier, the lowest share in the tracker so far. Where a notice gave a figure, the headline numbers add up to about 100.3 million, more than June, July and August together, and two parcel carriers, a car-sharing service and an image-sharing app account for 88.5% of it.

Who should act. Large consumer brands with customer portals, anyone running remote-access (VPN) devices facing the internet, and every organisation whose contractors log in to its systems.

One first action. List every internet-facing device and system, starting with remote-access gear, and patch those first, whatever severity rating the flaw carries.

Biggest uncertainty. How attackers got in. 52 of 91 (57.1%) of the distinct breaches do not say.

Posture

September was the month big consumer names arrived. The supplier waves of the summer faded, and in their place came large organisations breached in their own systems: Sagawa Express, Yamato Transport, Times Car and Gyazo. Government had its own warning in the Digital Agency, where a known flaw in a remote-access device let attackers in and contractors’ credentials did the rest. The last three days of the month, 24 notices, set up October.

The count

99 notices were disclosed between 1 and 30 September, up 5 from August’s 94 (5.3% more). The pace was steady at 16 to 21 a week until the last three days, which brought 24, including Sagawa Express, Yamato Transport and a hotel-booking platform.

Week Notices Through a supplier Own systems Supplier’s own notice Unclear
1–6 Sep 18 2 of 18 (11.1%) 15 1 0
7–13 Sep 16 1 of 16 (6.2%) 13 1 1
14–20 Sep 21 5 of 21 (23.8%) 14 2 0
21–27 Sep 20 6 of 20 (30.0%) 13 1 0
28–30 Sep 24 9 of 24 (37.5%) 13 2 0
September 99 23 of 99 (23.2%) 68 7 1

Weeks run Monday to Sunday and are cut at the month’s edges. “Own systems” means the organisation’s own systems were breached. “Supplier’s own notice” is a supplier announcing its own breach.

Supplier notices climbed through the month, from 2 in the first week to 9 in the last three days. That turned out to be the leading edge of October.

Notices vs distinct breaches

When one supplier is breached, every customer that publishes counts as a notice, but it is still one breach. To count breaches, I collapsed every notice linked to a named supplier into that supplier’s breach, and counted each own-system incident once.

About 91 distinct breaches produced the 99 September notices. 14 named supplier breaches account for 22 notices. 68 were breaches of the organisation’s own systems, 1 is unclear, and 8 came through a supplier the notice does not name. Those 8 are counted one each, so 91 is an upper bound.

9 of 14 (64.3%) named supplier breaches produced a single notice. No supplier produced more than 3.

Supplier breach What it does September notices Share of 99
ApplyNow Recruitment and applicant-management platform 3 3 of 99 (3.0%)
ShopServe (E-Store) Hosted online-shop platform (August breach, late customer notices) 3 3 of 99 (3.0%)
Temairazu Booking and room-inventory system for hotels and inns 3 3 of 99 (3.0%)
LEAN BODY (Metabase) Fitness-video service; breached analytics tool 2 2 of 99 (2.0%)
Studio No-code website builder 2 2 of 99 (2.0%)

Counts include the supplier’s own notice where it published one. Nine other supplier breaches produced one notice each.

Studio is worth a second look. Its two September notices were the first of 64 it produced in October.

Through a supplier

23 of 99 (23.2%) notices came from organisations hit through a supplier, and another 7 were suppliers announcing their own breach. Of the 23, 15 (65.2%) name the supplier and 8 (34.8%) do not, the weakest naming of the summer.

Government notices split evenly. 4 of 8 (50.0%) came through a supplier, including Osaka City, where support-scam pop-ups on a contractor’s after-school programme computers exposed 2,272 people, and Ichinomiya City through ApplyNow. The Digital Agency’s breach, below, is recorded as unclear in the tracker, so it sits outside both counts.

Former suppliers came back to bite again:

  • Yoshinoya reported 5,366 part-time applicants’ data exposed at ApplyNow, a supplier it had already stopped using.
  • KING JIM and PFU published notices about ShopServe shops they had closed. PFU had left the platform in October 2025.
  • WECLE was reached through its partner LEAN BODY, whose analytics tool was breached, and Calbee through a partner of the PhotoGoods photo-print service.

Notable cases

  • Sagawa Express disclosed on 30 September that its online parcel-tracking inquiry service had been accessed. The count reached up to about 53.23 million waybill records, a gross figure that includes duplicates, with sender and recipient names, addresses and phone numbers for parcels between 21 June and 30 September. Tokyo Shoko Research calls it the largest leak at a listed company or subsidiary since it began counting in 2012. Online tracking stayed offline until 23 October.
  • Helpfeel, which runs the Gyazo screenshot-sharing app, reported a command-injection flaw in an image-upload server, a bug that lets an attacker run their own commands, used on 11–12 September. About 23.62 million user records were read, along with image metadata.
  • Times Car, Park24’s car-sharing service, reported 6,628,007 accounts with signs of data being sent out. By its fifth report, ID-document images, such as driving licences, for 1,464,800 customers showed traces of access and transmission. Park24 offered ¥3,000 to each and expects a one-time loss of about ¥5.5 billion. That is what a breach costs when it lands on identity documents.
  • Yamato Transport confirmed about 5.31 million records taken from its Kuroneko deferred-payment service, covering customers back to October 2018.
  • The Digital Agency announced on 11 September that attackers used a known flaw in a remote-maintenance VPN device, rated medium severity and waiting in the patch queue, then used outsourced maintenance staff’s credentials to pull files in bulk. About 246,000 people were affected, mostly public servants and contractors. It detected the bulk access on 25 June, 78 days before the announcement.
  • Rohto Pharmaceutical reported an abused account on its mail-order telephone system, used to reach customer records. Its later count was 2,797,545 people, and call recordings for 9,715 of them included card numbers and expiry dates.
  • Temairazu, a booking system for hotels and inns, reported up to about 4.45 million guests exposed. It came to light when hotels reported suspicious messages being sent to their guests.
  • ApplyNow, a recruitment platform, reported a flaw in a data-analysis tool and about 390,000 people affected.

How attackers got in

The route is stated for 39 of 91 (42.9%) of the month’s distinct breaches. 52 of 91 (57.1%) say “unauthorized access” and nothing more. The shares below use only the 39 that state a route, so treat them as a sketch.

Route, in plain English Distinct breaches Closest MITRE ATT&CK technique
Misused account: stolen or guessed passwords, hijacked email or admin accounts 17 of 39 (43.6%) Valid Accounts (T1078)
Flaw in an internet-facing system or app: an unpatched or vulnerable website, server, plugin or programming interface (API) 13 of 39 (33.3%) Exploit Public-Facing Application (T1190)
Ransomware: files encrypted, often with data taken first 7 of 39 (17.9%) Data Encrypted for Impact (T1486), which describes the damage, not the way in
Other: malware on a contractor’s computer, payment fraud, supplier outages 2 of 39 (5.1%) Varies

MITRE ATT&CK is a public catalogue of attacker techniques that security teams use as a shared vocabulary. Supplier breaches take the route stated in the supplier’s own notice.

Misused accounts lead again, but flaws in internet-facing systems jump to a third of stated routes: an image server at Gyazo, an analytics tool at ApplyNow and LEAN BODY, a VPN device at the Digital Agency. Several of the biggest September breaches, Sagawa, Yamato and Times Car among them, say only “unauthorized access”.

What’s affected

78 of 99 (78.8%) notices list at least one type of data. The shares below use those 78 as the base.

Data type Notices Share
Email addresses 27 27 of 78 (34.6%)
Names 30 30 of 78 (38.5%)
Phone numbers 22 22 of 78 (28.2%)
Postal addresses 22 22 of 78 (28.2%)
Date of birth 7 7 of 78 (9.0%)
Passwords (hashed or not) 12 12 of 78 (15.4%)
Payment card data (including partial numbers) 3 3 of 78 (3.8%)

Data types are matched on the wording of each notice, and a notice that says “no card data” is not counted as card data. Passwords use the tracker’s own password flag.

Contact details lead, as always. The data that hurts most this month is not on the list: ID-document images at Times Car and card details in call recordings at Rohto. Passwords appear in 12 of 78 (15.4%) notices, including Gyazo and ShopServe’s late shops.

Records

September’s deduplicated total is 100,332,561. Notices whose figures already sit inside a supplier’s own total are left out: 10 of 99 (10.1%). That way each supplier breach counts once. Of the 89 that remain, 48 of 89 (53.9%) give a number.

The unit is the catch. Notices count email addresses, people, accounts or simply “records”, and the same person can appear in many of them. Read this as the sum of headline figures, not as a count of people.

Unit stated Records Share of total
Records, unit unspecified 89,567,583 89.3%
People 8,654,724 8.6%
Unit not disclosed 1,609,027 1.6%
Accounts and form submissions 440,427 0.4%
Email addresses 60,800 0.1%

Sagawa’s 53,230,000 records are 53.1% of the total, and Gyazo’s 23,620,000 another 23.5%. The top five make up 92.9%. Half the notices with a figure are below about 2,600.

The Sagawa figure counts waybills, so the same person can appear many times. That is the clearest reason to read this total as a sum of headline figures, not a count of people.

Organisation Headline figure Breached
Sagawa Express Up to about 53,230,000 waybill records (gross) Supplier’s own systems (shippers published too)
Helpfeel (Gyazo) About 23,620,000 user records Own systems
Times Car (Park24) 6,628,007 accounts Own systems
Yamato Transport About 5,310,000 records Own systems
Temairazu Up to about 4,450,000 people Supplier’s own systems

Disclosure speed

Only notices with an exact discovery date and an exact disclosure date are used: 28 of 99 (28.3%). Discovery here means when the organisation says it found out, which for a supplier’s customer is often when the supplier told it.

Group Notices with both dates Median days, discovery to disclosure Disclosed within 7 days
All 28 4 24 of 28 (85.7%)
Own systems 21 4 17 of 21 (81.0%)
Through a supplier 4 2 4 of 4 (100.0%)
Supplier’s own notice 3 4 3 of 3 (100.0%)

Disclosure is getting faster. 3 of 28 (10.7%) disclosed on the day they found out, and 1 of 28 (3.6%) took more than 30 days, 171 days at a company hit by malware. Supplier customers moved quickly this month, a median of 2 days across 4 notices.

The Digital Agency is not in the table because its notice gives two discovery dates. Either way, it took 78 days from detecting the bulk access to telling the public.

Sectors two ways

Counting notices and counting breaches give different pictures. The second column counts breaches of the organisation’s own systems, including suppliers that announced breaches of their own systems (75 in September).

Sector Notices Own-system breaches
Manufacturing / construction / utilities 17 of 99 (17.2%) 11 of 75 (14.7%)
IT / internet / software services 15 of 99 (15.2%) 13 of 75 (17.3%)
Transport / logistics 12 of 99 (12.1%) 11 of 75 (14.7%)
Education / research 11 of 99 (11.1%) 11 of 75 (14.7%)
Government / public sector 8 of 99 (8.1%) 3 of 75 (4.0%)
Retail / e-commerce 8 of 99 (8.1%) 7 of 75 (9.3%)
Other 8 of 99 (8.1%) 6 of 75 (8.0%)
Travel / hospitality / leisure 7 of 99 (7.1%) 3 of 75 (4.0%)
Media / entertainment 6 of 99 (6.1%) 5 of 75 (6.7%)
Finance / insurance 4 of 99 (4.0%) 4 of 75 (5.3%)
Healthcare / pharma 2 of 99 (2.0%) 1 of 75 (1.3%)
Restaurant / food service 1 of 99 (1.0%) 0 of 75 (0.0%)

Manufacturing, construction and utilities top the notice count with 17 of 99 (17.2%). Transport and logistics, usually a small slice, produced 12 of 99 (12.1%): Sagawa, Yamato, Japan Post twice, Tokyo Metro, Times Car and Keio among them. By own-system breaches, IT and internet companies lead with 13 of 75 (17.3%).

Outside counts

The Personal Information Protection Commission (PPC) received 303 initial reports of personal-data leaks from unauthorized access in September, up from 235 in August. That was the highest month in the PPC’s table, which starts in April, until October, which reached 830 in its first 16 days. It counts reports filed with it rather than public notices, and leaves out financial and telecom firms and public bodies.

Tokyo Shoko Research (TSR), which counts only listed companies and their subsidiaries, says its monthly count stayed in the teens through September before jumping to 109 in October.

One practical tip

List every device and system that faces the internet, and patch those first. Remote-access gear, such as VPNs, firewalls and remote-maintenance boxes, comes top. A “medium” rating on a device anyone on the internet can reach should be treated as urgent: the Digital Agency’s flaw was rated medium and was waiting its turn.

Then make sure contractor accounts on those systems use phishing-resistant MFA and are switched off when the work ends.

30, 60 and 90 days

Review every supplier that touches personal data or logs in to your systems. Keep the ones that are necessary, and bring the rest in-house or drop them. Fewer contractors with remote access means fewer accounts to secure and fewer ways in: operational efficiency as well as lower risk.

When Action Owner Done when
30 days List every internet-facing device and system, with its owner and patch status. IT operations One list; remote-access devices flagged.
60 days Patch or replace everything internet-facing that is behind, whatever the severity rating. Phishing-resistant MFA on every contractor and remote-maintenance account. IT operations, with suppliers No overdue patches on internet-facing kit; contractor accounts covered.
90 days Review every supplier with remote access. Keep the necessary ones, end access for the rest, and add a clause requiring them to tell you of a breach within a set number of hours. Head of IT or security, with legal and procurement Each supplier has a decision; clauses in critical contracts.

What remains unknown

  • How most attackers got in. 52 of 91 breaches (57.1%) do not say, including several of the biggest.
  • How many people were really affected. Units are mixed, the largest figure counts waybills, and 41 of the 89 counted notices (46.1%) give no number.
  • Which customers of Sagawa, Temairazu and ApplyNow will still publish. Many did so in October.
  • How complete September is. These rows were added later, so small notices may be missing.

How we counted

Each row of the Japan Cybersecurity Tracker is one organisation’s published notice about a cyber incident in Japan. This briefing uses rows whose disclosure date falls between 1 and 30 September 2026, from the snapshot of 16 November 2026. The June to September rows were mostly added after the fact, in a backfill, so small notices that never reached the press are more likely to be missing than in later months.

  • Included: incidents disclosed by the organisation itself, or a supplier’s notice confirmed by the customer’s own notice. Hijacked email and admin accounts count even when no leak is confirmed.
  • Not included: ransomware gangs’ claims on the dark web without a notice, lost USB sticks and other physical losses, and precautionary shutdowns with no incident.
  • “No leak confirmed” is recorded as stated. It is not read as “no leak”.
  • Supplier links: of the 23 notices through a supplier, 15 (65.2%) name the supplier and 8 (34.8%) do not name one.
  • Distinct breaches: notices are collapsed to the parent breached supplier named in the data. Own-system incidents and unnamed-supplier notices count one each, so the distinct count is an upper bound.
  • Records: the headline or maximum figure each notice gives, including later revisions. Blank stays blank, not zero. Figures already inside a supplier’s total are excluded from the sum.
  • Dates: notices are placed by disclosure date. 13 tracker rows have no exact disclosure date. None of them is placed in this month, and they are left out of every monthly figure.
  • Limits: categories come from the wording of notices, much of it translated from Japanese. The data is a snapshot and changes as reports are revised.

The full row-level dataset is available on request through the dataset page, which also explains the method and how to cite it.

If you need any help, don’t hesitate to get in touch.

Matt Ketchum