From July 30 to November 12, a third party repeatedly opened files on the service Japan's ministries use to share meeting papers. Access began 21 days after the Digital Agency contained an earlier intrusion there. Intruders read Bulgaria's document exchange for five years, taking over 62 million pages. A suspected Chinese-nexus threat actor's artificial intelligence (AI) agents rooted a university imaging server through an unknown flaw. Malware on home computers waits for military identity cards, and Iran-linked hackers impersonated a real expert to reach an Australian broadcaster.
An Unknown Party Accessed Japan's Shared Ministry Files From July 30 to November 12
Ministry meeting papers on Japan's Government Solution Service (GSS) were opened multiple times by someone outside the authorized sharing partners before the Digital Agency detected the access on November 12, the agency disclosed on November 20 and Nikkei reports. GSS supplies the shared laptops, network, and security used by about 150,000 people at 25 government bodies, Asahi Shimbun reports. Its sharing service carries meeting materials and other large files among ministries and out to businesses. Quotations here are translated from Japanese.
The agency immediately isolated the service from external networks and took it offline the same day. Because the files are presumed to hold personal data, the agency reported to the Personal Information Protection Commission (PPC) and is investigating with police and outside specialists. The notice names no affected ministry. The service runs on a commercial cloud platform the agency has not named, also used to exchange files with local governments, Nikkei reports.
Users include the Ministry of Internal Affairs and Communications (MIC), the Ministry of Economy, Trade and Industry (METI), the Ministry of Justice (MOJ), the Financial Services Agency (FSA), and the Nuclear Regulation Authority (NRA), according to Nikkei. GSS may hold material up to confidentiality level 2 of 3, "not secret documents, but not information disclosed to the public," Asahi Shimbun reports. The Ministry of Foreign Affairs (MOFA) and Ministry of Defense (MOD) do not use GSS, and the NRA keeps nuclear material protection data separate.
CNET Japan notes the Digital Agency has published no file count, no figure for people affected, and no access method. Officials believe GSS security itself "had no problem" this time, Yomiuri Shimbun reports. If that assessment holds, the access may have come through a valid account, a compromised endpoint, a sharing partner, or the cloud provider.
In the first GSS breach disclosed this autumn, a third party exploited a vulnerability in a virtual private network (VPN) device, the September 11 notice says. The intruder then used a maintenance account to access large numbers of files, which the agency detected on June 25. The agency contained the intrusion on July 9 by suspending the account and isolating the compromised device from external networks. About 246,000 records may have leaked, 189,000 on government staff and 57,000 on outside businesses and individuals.
Those records came from user registration forms and cover company staff and web meeting participants, the agency's Q&A says. The VPN flaw was public before the attack and rated Medium. Then Digital Minister Hisashi Matsumoto said patching was queued by severity when the intruder gained access. He withheld the ministries' names so "the contents of government as a whole" would not become known. The later access began 21 days after that containment.
MOFA detected unauthorized access on July 27 to its network for exchanges with outside parties, three days before the access to ministry files began. About 3,600 personal records may have leaked there, about 900 of them on people outside the ministry.
The September intruder may have returned by phishing the people it catalogued, since the registration data gave it names and work email addresses for businesses and web meeting participants. Two other explanations fit the published facts. A separate compromise at one sharing partner may have exposed that partner's files. That was the pattern of the 2021 Fujitsu ProjectWEB breach, in which the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) lost at least 76,000 email addresses. Or a threat actor compromised the cloud provider itself.
ProjectWEB also leaked system details and study papers from the National center of Incident readiness and Strategy for Cybersecurity (NISC) and MOFA. Three tests would separate the explanations. Did the access use valid partner accounts, do the opened files concentrate in one partner's shares, and does the infrastructure match September's?
The National Police Agency (NPA) and NISC attributed campaigns against Japan since 2019 to MirrorFace (aka Earth Kasha) in January 2025. The agencies suspect Chinese involvement, aimed mainly at stealing national security and advanced technology information. The NPA describes three campaigns. From December 2019 to July 2023, MirrorFace emailed LODEINFO malware to think tanks, serving and retired government officials, politicians, and the media. From February to October 2023, the threat actor exploited VPN devices and stolen credentials against semiconductor, manufacturing, communications, academic, and aerospace targets.
The NPA says MirrorFace exploited flaws in Array Networks Array AG (CVE-2023-28461), Fortinet FortiOS (CVE-2023-27997), and Citrix ADC (CVE-2023-3519). Since June 2024 MirrorFace has emailed links to ANEL malware from Gmail, Outlook, and hijacked legitimate addresses, resending stolen past correspondence with small changes so the messages read naturally. That tradecraft matches the September entry through a VPN device and the phishing route the first explanation proposes, but matching tradecraft identifies no threat actor.
Chinese state-backed hackers were also suspected in the NISC email breach disclosed in 2023, which Beijing denied while blaming the United States, The Record noted from Financial Times reporting.
No public evidence ties MirrorFace or the NISC intruder to GSS, and no one has named a threat actor for the access. Repeated access from July 30 to November 12 to meeting files shared with companies and local governments, with no extortion described, followed a breach that harvested the platform's user directory. I assess with moderate confidence the access served intelligence collection, and with low confidence the September intruder returned.
Withholding the affected ministries keeps nothing from the intruder, who knows what was opened. The list is missing only for the businesses and local governments sharing files with ministries. Prime Minister Sanae Takaichi announced a ¥10 billion emergency cyber package on November 6, and Japan's breach disclosures now include three against central government systems since September.
Intruders Read Bulgaria's Document Exchange for Five Years
Intruders held hidden access to Bulgaria's government networks from 2021 and took at least 130 gigabytes, over 62 million pages, through the infrastructure institutions use to exchange documents. Minister of Innovation and Digital Transformation Ivan Vasilev gave the figures at a Council of Ministers briefing on November 18, Mediapool reports. The stolen material covers Council of Ministers work, reports, and correspondence between ministers. The ministry first disclosed the access on August 5. Quotations here are translated from Bulgarian.
The alarm came on June 23, when monitoring software saw an employee "still working and processing information" while away from work, national cyber security coordinator Velizar Varbanov told Fakti. The intruders gained initial access through phishing, exploited vulnerabilities, and compromised accounts. They infected several hundred computers, servers, and virtual machines. Information Services, the state integrator, announced in 2024 that the US Trade and Development Agency (USTDA) would help build a national cyber security system. No source ties that system to the June alarm.
Vasilev says "several international hacker groups" were behind the theft. He refuses to name a country and says it is not established whether they worked for foreign intelligence services. Varbanov says the servers included compromised machines in many countries, so their locations prove nothing. He says the aim was to anticipate "the state's next moves." Opposition lawmaker Bozhidar Bozhanov, a former e-government minister, named China, Russia, Iran, or North Korea as likely sponsors.
The referral to prosecutors asks about the "actions and inactions of state bodies," including whether gaps were deliberate or the intruders had help. Rumen Radev's government took office on May 8, 2026, after most of the five-year compromise had already occurred, so the inquiry falls mainly on its predecessors. Declining to name a state also avoids a diplomatic cost. Neither choice changes the theft, and "several groups" could describe one intelligence service using varied infrastructure.
Bulgaria supplied about a third of the ammunition Ukraine needed early in the war, former Prime Minister Kiril Petkov said, and its prosecutor traced a 2022 attack on the presidency to Magnitogorsk. I assess with moderate confidence the theft was state intelligence collection, matching the pattern jark.me analyzed when Russia's Federal Security Service (FSB) read Romania's government mail. The public evidence does not support naming the state.
An interagency group isolated the affected environment, tightened access to critical systems, cut the intruders' connection to their infrastructure, and removed them from state systems, working with international partners including the European Union Agency for Cybersecurity (ENISA), Vasilev said, according to Mediapool and BNT. He says no evidence so far shows Bulgarian citizens' personal data was affected.
Bulgaria is adding multifactor authentication and administrator account controls and extending monitoring to 28 municipal and regional administrations, Vasilev told BNT. The national cyber security strategy, overdue for years, is still not adopted.
Cairn Agents Rooted a Medical Imaging Server in East Asia
A suspected Chinese-nexus threat actor's AI agents compromised a university's medical imaging server through an unknown path traversal flaw, going from mapping its interface to writing files in about 11 minutes. The campaign ran about 48 hours from September 8 against universities in East Asia outside mainland China, Amazon Threat Intelligence reported on November 20. Amazon traced the campaign on September 10 to the threat actor's own unauthenticated Cairn control panel.
Cairn, an open source autonomous penetration testing framework, placed third of 610 teams at a Tencent Cloud hacking contest, the project documentation says. The agents' single goal was "obtain shell access." Amazon says they spent 16 hours and 148 actions on an open Redis database and tried more than 30,000 credential pairs on a graphics processing server, both without success. On MONAI Label, an open source tool for building medical AI annotation models, they wrote a Python reverse shell into sitecustomize.py and gained root privileges.
The threat actor then handed the agents a proxy into a second university running MONAI Label. The fix merged on October 19 and sat public for 30 days before the advisory of November 18, which listed no patched version. Amazon wrote on November 20 that maintainers "released a fix," yet MONAI Label 0.8.6 shipped only on November 22. The release fixes CVE-2026-133968, which MITRE scores 9.8 in the Common Vulnerabilities and Exposures (CVE) record and the National Vulnerability Database (NVD) lists as exploited since September.
Amazon's moderate-confidence attribution rests on Simplified Chinese prompts, China Standard Time timestamps, and targeting. Palo Alto Networks Unit 42 tracks another Cairn user, CL-STA-1238, which ran 25 attacks from 154.44.31[.]4 in Hong Kong, as jark.me analyzed. Amazon lists 142.4.98[.]145 and 142.4.107[.]34, and Gambit Security found an eCrime threat actor using the framework against at least 27 retailers.
The framework now serves both espionage and theft, so its presence in an intrusion names no threat actor. A compromised graphics processing server is worth taking for its computing power alone, and an open control panel is poor tradecraft for an intelligence service. Administrators of MONAI Label servers should install 0.8.6, enable authentication, which the release notes say is off by default, and block internet access to the servers.
Malware Hijacks Military Identity Cards Left in Home Card Readers
Malware on personal computers records the personal identification number (PIN) a cardholder types for a Common Access Card (CAC). The malware then uses the inserted card to sign in to federal and Department of War (DoW) websites for as long as the card stays in the reader. A National Security Agency (NSA) official warned the Department of War news service on November 16. The official said the threat "is being used by bad actors." Personnel typically use home readers for work email after hours.
The NSA names no threat actor, no malware, and no victim count. The official described three infection routes. Threat actors may target individual employees with phishing, compromise sites they visit as watering holes, or infect "many computers or home network devices" and act only when one visits a department site. The Department of War name comes from Executive Order 14347.
AlienVault documented a Sykipot variant keylogging smart card PINs on computers running ActivClient in January 2012, The Register reported, and SecurityWeek wrote Chinese hackers were said to be responsible. The 2026 warning adds the third route, which opens the door to access brokers. An eCrime botnet could infect machines indiscriminately, find those holding active military credentials, and sell the sessions without choosing the cardholders beforehand, all from personal computers the department does not manage.
The NSA tells cardholders to use a reader on as few computers as possible, pull the card whenever stepping away, and keep the operating system, applications, and security software patched.
Iran-Linked Hackers Targeted Three Australian Critics of Tehran
Iran-linked hackers posing as a real United States think tank expert approached broadcaster Erin Molan, she told Iran International on November 21. She was one of three Australians targeted, along with journalist Sharri Markson and political commentator Michael Kroger. She says the head of the Australian Security Intelligence Organisation (ASIO) gave her the count. The first message came on X, quoting her support for the Iranian people and her late father, Jim Molan, a retired major general and senator.
A month later Molan clicked a meeting link, the screen behaved oddly, and nobody joined. She found the missed security alert only after reading Markson's account. "The person inside her account was not her," Molan said of the expert.
Markson's approach began in August through the X account of a former Trump administration official. The account offered Hudson Institute researchers and a contact called "Cathleen Batz." An October 8 video call would not connect, and by then the threat actor had allegedly accessed her Google Drive. News Corp's own experts attributed the approach to CHARMING KITTEN (aka Mint Sandstorm), The Nightly reports. Markson exposed the lure by checking with her contact on WhatsApp and Signal, and he had "no idea" what she meant.
The threat actor controlled the official's account through X's delegate feature, J-Wire reports. Mike Burgess, director-general of ASIO, named CHARMING KITTEN to Markson himself, and his service found Kroger on an Iranian target list.
Microsoft reported in January 2024 that a Mint Sandstorm subgroup, overlapping CHARMING KITTEN, built rapport with benign emails and sent lures from "legitimate but compromised" accounts. Google Threat Intelligence Group (GTIG) documented APT42 posing as journalists and event organizers.
Australia expelled Iran's ambassador on August 26, 2025 and listed the Islamic Revolutionary Guard Corps (IRGC) as a state sponsor of terrorism that November. In June Burgess said an Iran-based Australian citizen, a senior Qods Force agent, orchestrated the October 2024 firebombing of the Lewis Continental Kitchen in Bondi. He says Iran could "conduct assassinations on Australian soil." Access to an inbox gives that service contacts, schedules, and movements it can use for physical targeting.
I assess with moderate confidence the targeting of Molan, Markson, and Kroger belongs to one Iranian intelligence collection effort against Australian critics of Tehran. The Molan and Markson approaches share the think tank pretext and hijacked real accounts, Kroger sits on the target list Burgess described, and the IRGC link rests on the Markson case.
Analyst Note
Each intrusion crossed a boundary where no single defender held the full record. The GSS sharing service sits between ministries, local governments, and companies, and Bulgaria's exchange links institutions across the government. The CAC malware runs on home computers the Department of War does not manage. Cairn worked from a control panel the universities never saw, and Iran's operatives reached Markson through an X account belonging to someone else.
Bulgaria's alarm was an employee account active after hours, and Japan detected the September intrusion when a maintenance account accessed large numbers of files. Amazon spotted the Cairn campaign only because the threat actor left its panel open, and Molan learned of her breach only after reading about Markson. Every detection described here came through odd behavior or an outsider.
MONAI Label required no login by default, so the Cairn agents needed no account to steal, and watching accounts would have shown nothing. Only authentication and keeping the server off the internet would have stopped them.
The Digital Agency withholds the affected ministries, Bulgaria's government will not name a state, and the NSA names no malware, while the head of ASIO named the threat actor to Markson directly. The silence protects no secret from the intruders, who know what they took. It leaves the partners, cardholders, and administrators who could act without the detail they need.
Patch edge devices by exposure rather than severity score, and watch valid accounts for work at odd hours and odd volumes. Switch on authentication wherever a tool ships without it, keep identity cards out of home computers, and confirm any approach through a second channel before clicking.
Analyst Annex
Evidence status as of November 22, 2026. The records cover five items, each kept as a separate record set. They are the unauthorized access to shared ministry files on Japan's Government Solution Service (GSS) with the earlier GSS breach, the five-year intrusion into Bulgarian government networks, the Cairn agent campaign against universities in East Asia, the National Security Agency (NSA) warning on Common Access Card (CAC) malware, and the Iranian approaches to Australian commentators.
Indicators come only from Amazon, the one source publishing indicators for these items. The ATT&CK mapping covers the procedures each source describes. No source describes how the July to November GSS access worked, so that item has no mapped technique.
Confirmed marks evidence published by a victim, vendor, regulator, or government authority about its own systems, decisions, or investigation. Reported marks technical findings or attribution assessments published by a third party. Claimed marks statements by a seller or self-identified threat actor that remain unverified. Provisional marks an analytical mapping supported by reported behavior but not made by the original source.
Indicators
| Indicator | Context | Status |
|---|---|---|
| IP Addresses | ||
142.4.98[.]145 |
Address Amazon lists in records dated September 8 to 10, 2026 for the Cairn campaign Source: Amazon | Reported |
142.4.107[.]34 |
Address Amazon lists in records dated September 8 to 10, 2026 for the Cairn campaign Source: Amazon | Reported |
MITRE ATT&CK Mapping
| Tactic | Technique | Procedure | Status |
|---|---|---|---|
| GSS Breach Through a VPN Device, May to July 2026 | |||
| Initial Access | T1190 Exploit Public-Facing Application | A third party entered GSS through a vulnerability in a VPN device Source: Digital Agency, MITRE ATT&CK | Provisional |
| Persistence | T1078 Valid Accounts | A maintenance account was used to access a large number of files on a server Source: Digital Agency, MITRE ATT&CK | Provisional |
| Intrusion Into Bulgarian Government Networks, 2021 to 2026 | |||
| Initial Access | T1566 Phishing | Varbanov lists phishing among the entry methods Source: Fakti, MITRE ATT&CK | Provisional |
| Initial Access | T1190 Exploit Public-Facing Application | Varbanov lists system vulnerabilities among the entry methods Source: Fakti, MITRE ATT&CK | Provisional |
| Persistence | T1078 Valid Accounts | Compromised accounts, with an employee account active after working hours setting off the alarm Source: Fakti, MITRE ATT&CK | Provisional |
| Cairn Agent Campaign Against Universities in East Asia, September 8 to 10, 2026 | |||
| Reconnaissance | T1589.003 Gather Victim Identity Information: Employee Names | Agents built about 80 real usernames from faculty directories, lab websites, and papers Source: Amazon, MITRE ATT&CK | Provisional |
| Credential Access | T1110.003 Brute Force: Password Spraying | Credential sprays of more than 30,000 combinations against SSH on a university GPU server and against PostgreSQL elsewhere Source: Amazon, MITRE ATT&CK | Provisional |
| Discovery | T1046 Network Service Discovery | Agents mapped the internal Docker network topology through an SSRF flaw in a co-located Qdrant instance Source: Amazon, MITRE ATT&CK | Provisional |
| Initial Access | T1190 Exploit Public-Facing Application | Path traversal in the MONAI Label upload endpoint, exploited in about 11 minutes Source: Amazon, MITRE ATT&CK | Provisional |
| Execution | T1059.006 Command and Scripting Interpreter: Python | Python reverse shell written to sitecustomize.py, running as root Source: Amazon, MITRE ATT&CK | Provisional |
| Command and Control | T1090 Proxy | Proxy tunneling into a second university network Source: Amazon, MITRE ATT&CK | Provisional |
| CAC Malware on Personal Computers, warned November 16, 2026 | |||
| Collection | T1056.001 Input Capture: Keylogging | Malware records the CAC PIN on personal computers Source: Department of War, MITRE ATT&CK | Provisional |
| Iranian Approaches to Australian Commentators, August to November 2026 | |||
| Resource Development | T1586.001 Compromise Accounts: Social Media Accounts | An official's account infiltrated through the platform delegate feature to approach Markson Source: J-Wire, MITRE ATT&CK | Provisional |
| Initial Access | T1566.003 Phishing: Spearphishing via Service | Approach on X with a meeting link, a month after first contact Source: Iran International, MITRE ATT&CK | Provisional |
| Stealth | T1684.001 Social Engineering: Impersonation | Impersonated a real think tank expert to Molan and wrote to Markson through a real former official's account Source: Iran International, J-Wire, MITRE ATT&CK | Provisional |
| Collection | T1530 Data from Cloud Storage | The threat actor allegedly reached Markson's Google Drive Source: The Nightly, MITRE ATT&CK | Provisional |
Entities
| Entity | Role | Status |
|---|---|---|
| Threat Actors | ||
| Unidentified Threat Actor (GSS, July to November 2026)Unattributed | Accessed shared ministry files on GSS from July 30 to November 12, 2026. No threat actor named Source: Digital Agency | Confirmed |
| Unidentified Threat Actors (Bulgaria)Unattributed, several groups per the minister | Vasilev refused to name countries and says ties to foreign intelligence services are not established Source: Mediapool | Reported |
| Unidentified Chinese-Nexus Threat ActorSuspected cyber espionage | Amazon assesses with moderate confidence a Chinese-nexus threat actor ran the Cairn campaign Source: Amazon | Reported |
| CHARMING KITTEN (aka Mint Sandstorm)Iran, linked to the Islamic Revolutionary Guard Corps (IRGC) | Named for the Markson approach by News Corp experts and, per J-Wire, by Burgess. Not established for every Australian target Source: The Nightly, J-Wire | Reported |
| Organizations and Victims | ||
| Digital AgencyJapanese government agency operating GSS | Isolated the GSS sharing service from external networks and took it offline on November 12, 2026 Source: Digital Agency | Confirmed |
| Ministry of Foreign Affairs (MOFA)Japanese ministry outside GSS | Detected unauthorized access to its external network on July 27, about 3,600 records Source: Ministry of Foreign Affairs | Confirmed |
| Bulgarian state administrationGovernment networks | At least 130 gigabytes, over 62 million pages, taken from 2021 Source: Mediapool | Reported |
| Universities in East AsiaOutside mainland China | MONAI Label server rooted, and a second university reached by proxy Source: Amazon | Confirmed |
| Erin MolanAustralian broadcaster | Approached on X by a sender using a real think tank expert's name Source: Iran International | Reported |
| Sharri MarksonAustralian journalist | Google Drive allegedly accessed after a two-month approach Source: The Nightly | Reported |
| Michael KrogerAustralian political commentator | On an Iranian target list, warned before the suspected operatives could compromise his accounts Source: The Nightly | Reported |
| Tools and Services | ||
| CairnOpen source autonomous penetration testing framework | Used by the Chinese-nexus threat actor against universities in East Asia Source: Amazon | Confirmed |
| Vulnerabilities | ||
| CVE-2026-133968MONAI Label 0.8.5 and earlier, scored 9.8 by MITRE | Path traversal fixed in MONAI Label 0.8.6, released November 22, 2026 Source: MONAI Label, NVD | Confirmed |
Related Activity
The records below cover earlier incidents linked to the items. None was observed in the incidents covered here.
| Item | Detail | Status |
|---|---|---|
| Fujitsu ProjectWEB BreachJapanese supplier breach, May 2021 | At least about 76,000 email addresses and system information leaked from the transport ministry Source: Ministry of Land, Infrastructure, Transport and Tourism | Confirmed |
| MirrorFace (aka Earth Kasha) CampaignsSuspected Chinese involvement, since about 2019 | The National Police Agency (NPA) describes LODEINFO and ANEL phishing against think tanks, government, politicians, and media, and VPN device exploitation of CVE-2023-28461, CVE-2023-27997, and CVE-2023-3519 against semiconductor, manufacturing, and aerospace targets Source: National Police Agency | Confirmed |
| CL-STA-1238Cairn campaign, August to September 2026 | Unit 42 tracked 25 Cairn attacks against at least 19 organizations from a Hong Kong server Source: Unit 42 | Reported |
| Cairn eCrime CampaignFinancially motivated, from July 2026 | Gambit Security found at least 27 companies compromised between 10 and 15 September alone Source: Gambit Security | Reported |
| Sykipot Smart Card VariantSmart card PIN theft, January 2012 | A Sykipot variant targeted PCs with smart card readers running ActivClient and keylogged PINs Source: The Register | Reported |
| Mint Sandstorm CampaignIranian social engineering, January 2024 | Sent lures from legitimate but compromised accounts after building rapport Source: Microsoft | Confirmed |
| VENOMOUS BEAR in RomaniaEarlier jark.me analysis | FSB collection on Romania's dealings with NATO and the EU through a government mail server Source: jark.me | Reported |
Intelligence Gaps
| Gap | Why It Matters | What Would Close It |
|---|---|---|
| GSS access method | The path could run through a valid account, a compromised endpoint, a sharing partner, or the cloud provider, and the Digital Agency has published no access method | The Digital Agency investigation results or the cloud provider's logs |
| Link to the September GSS intruder | The later access began 21 days after the July 9 containment, and no infrastructure has been compared publicly | Matching infrastructure or accounts between the two GSS incidents |
| Bulgarian sponsor | Vasilev refused to name a state, and server locations include compromised machines | A government attribution or published indicators |
| Cairn campaign sponsor | Amazon's moderate confidence rests on language, time zone, and targeting | Infrastructure overlap with CL-STA-1238 or another tracked threat actor |
| CAC malware family and scale | The NSA names no malware, threat actor, or victim count | An NSA advisory or vendor report naming the malware |
| Molan compromise outcome | The odd meeting screen and missed alert do not show which account, device, or data was reached | Platform logs, device forensics, or an account from Molan or the government of what was affected |
| Australian targeting scope | No public record shows whether one threat actor handled all three targets through shared infrastructure | A comparison of accounts, meeting links, and infrastructure across the Molan, Markson, and Kroger cases |
| Iranian attribution | The Nightly credits News Corp experts, while J-Wire credits Burgess | An official statement from the Australian Security Intelligence Organisation (ASIO) |