In the first days of October 2026, several Japanese government bodies issued warnings in quick succession, including IPA (the Information-technology Promotion Agency), the Ministry of Economy, Trade and Industry (METI), the National Cybersecurity Office and the Financial Services Agency (piyolog’s compilation). The reason: data breaches from unauthorized access keep being disclosed across Japan. IPA’s notice asks the executives of companies holding large volumes of data to run an urgent review.
This isn’t a news recap. It’s an attempt to answer one question: what does a business leader actually have to decide? It covers newer ideas such as zero trust and AI security, and it leans on what the official documents say. Sources are given next to each statement, and this is a summary of public documents, not an individual assessment. Things are changing daily, so everything here is as of October 11, 2026.
According to IPA’s notice of October 9, breaches are being disclosed across a wide range of organizations, including financial institutions and telecom carriers. The pattern that stands out: online services and account systems holding large amounts of personal data are breached and the data is stolen.
The cause has not been pinned down to a single product’s vulnerability. But judging from the cases disclosed so far, IPA says the starting point appears to be either a compromised externally facing application or service, or a compromised account. Stolen data also tends to lead to secondary harm: misuse, extortion threats that hint at publishing it, and illegal sales (IPA’s notice).
Ransomware hasn’t gone away either. According to a report on the National Police Agency’s published figures, 2025 saw 226 reported ransomware cases in Japan, the second-highest ever. In IPA’s “10 Major Security Threats 2026,” ransomware ranks first for organizations, and attacks targeting the supply chain and contractors rank second.
What the big incidents of last year show isn’t flashy attack technique. It’s the unglamorous parts of defense. I’ll stick to what the companies disclosed and what has been reported and analyzed.
Askul was hit by ransomware on October 19, 2025. According to its investigation, the trigger was a leaked credential for an administrator account used by a contractor, and that account did not have multi-factor authentication. The attacker got in through the VPN with a stolen ID and password, disabled the endpoint monitoring software (EDR), then encrypted the logistics and other systems and deleted the backups. The company explained that there was no backup environment designed with ransomware in mind, which prolonged recovery (IT Pro (Impress), in Japanese).
Asahi Group Holdings was hit on September 29, 2025. At the press conference, the company explained that the attacker entered through a network device at a group site and then took administrator privileges at the data center. The company has not officially named the device used for entry. It took more than two months for part of the systems to come back (Trend Micro’s analysis, in Japanese).
What the two cases tell us is that defense depends less on advanced technology than on basic operations: authentication, privileges, detection and backups.
For a long time, corporate security worked like a castle wall. The inside of the company network is safe; you watch what comes in from outside. A VPN is the key to the gate.
But if a legitimate ID and password are stolen, that key opens easily. And once someone is inside, everything inside is trusted, so they can move around freely. The Askul case followed that pattern.
Zero trust is the shift in thinking. The US National Institute of Standards and Technology (NIST) document SP 800-207 sets out the idea that no implicit trust should be granted based solely on network location or on who owns an asset. Being inside the office, or using a company laptop, isn’t enough. You check, every time, who is trying to reach what, and from which device. Zero trust isn’t a product name. It’s that way of thinking.
Both Askul and Asahi have named zero-trust-based measures as part of their direction for preventing a repeat (IT Pro, Trend Micro).
A smaller company doesn’t need to buy a zero trust product tomorrow. In practice you can begin with four things:
One note: the word “zero trust” does not appear in IPA’s notice. But the measures it asks for (stronger authentication, a review of privileges, an inventory of accounts, a review of logs) point, as I read it, in much the same direction.
IPA says the starting point appears to be externally facing applications and services: customer membership sites, booking systems, app APIs, admin screens, contact forms.
Here is what the checks and measures in IPA’s notice look like in plain terms for a web team:
That last one pays off the most. Less data to steal means less damage. Even a routine that deletes old membership data and inquiry history after a set period reduces exposure.
In IPA’s 2026 list, “cyber risks from the use of AI” entered the organizational top 10 for the first time, at number three. But AI security means different things depending on where you stand.
The first is employees using AI. Paste customer data or internal documents into a tool like ChatGPT and that information may leave the company. Japan’s AI Guidelines for Business, issued by the Ministry of Internal Affairs and Communications and METI, added in the March 2026 update the importance of minimizing the data you input, to avoid unintended leakage. They also call for mechanisms that keep a human in the loop where outputs could have serious consequences. In practice it starts with finding out who is using which AI and how, then deciding which tools are allowed and what information must never go in.
The second is building AI into your own services, such as a support chatbot or an AI that searches internal documents. The classic risk here is prompt injection: a crafted input that makes the AI produce output it shouldn’t. The attacker may type it directly, or plant instructions in a web page or file the AI reads. The technical guidelines on AI security published by the Ministry of Internal Affairs and Communications on March 27, 2026 list checks on inputs, on externally referenced data and on outputs, plus log retention and usage limits. In OWASP’s Top 10 for LLM Applications (2025), prompt injection and sensitive information disclosure are the first two entries.
The third is AI agents: AI that reads email and operates systems on a person’s behalf. They are convenient, but the broader the privileges you give one, the broader the damage if it is hijacked. The ministry’s guidelines currently leave AI agents out of scope, saying the technology is still developing. METI has only reached the stage of announcing, in September 2026, a plan to compile good practices for companies that use AI agents; there is no formal guideline yet. That means each company has to protect itself. The basics: give an agent the minimum privileges, and have a person approve anything that can’t be undone, such as deletion or payment.
You might also wonder whether attackers are using AI. For this wave of incidents, some analyses raise the possibility, but the analysis itself, and a joint document by experts, say no traces directly showing AI use have been confirmed (piyolog’s summary). My own view is that, for now, the more accurate reading is not that AI has made attacks new, but that the basics are being broken.
Here is an order of priority, based on IPA’s notice.
This week
This month
This quarter
If your review turns up unusual logs or accounts you don’t recognize, IPA strongly recommends asking a specialist security vendor for a detailed investigation. Don’t just delete them and move on.
What IPA and METI both stress in these notices is that security should not be left to the person in charge; executives should lead (METI’s announcement is also reported by Nikkei xTECH). IPA points to the three principles for executives in METI’s Cybersecurity Management Guidelines. Treat cybersecurity risk as a key management issue. Look beyond your own walls to sites, partners and contractors. Communicate actively with stakeholders, in normal times and in emergencies.
It looks like a technical subject, but what has to be decided is a management matter. How much data do we hold? How much do we trust contractors with? How much privilege do we give an AI? Where do we put a human approval step? All of these are choices between risk and convenience, and no single engineer should be making them alone.
We do AI adoption and business systems work that takes us inside our clients’ operations. In that work, we see security as something to think through from the start, when deciding which processes get AI and where a person checks, not something bolted on afterwards. If you’d like help sorting out where you stand, get in touch below.
We are currently taking free consultations on security. Questions like "Is our site safe?" or "Where should we start checking?" are welcome.
43 W 33rd St, #202, New York, NY 10001
Phone: +1 (848) 281-7853
Email: info@supersonicdesign.com
