On 20 November 2026, Japan’s Digital Agency announced a possible second GSS data breach this year: files stored in the file-sharing service it provides through the Government Solution Service may have been leaked to outside parties. The agency says it detected possible access by someone other than the intended recipients on 12 November, cut the service’s external communications at once and suspended the service the same day. Its investigation has established that the access happened between 30 July and 12 November, a window of 105 days, and the details of that period are still being examined. The agency has not said how many files were involved, what they contained, which ministries were affected or how the intruder got in.
Eight days passed between detection and disclosure. In September the agency took 78 days to tell the public about anomalous file access it had detected, so the disclosure clock has sped up considerably.
Coverage usually lines the two disclosures up as a sequence of embarrassments. I find them more useful as a pair of engineering case studies, because they broke in different places for different reasons. September began at a network appliance and ended with a privileged account reading files in bulk, which makes it a story about vulnerability triage and privilege concentration. November happened inside a service whose whole job is handing files to people outside the organisation. An externally reachable sharing platform is a trust boundary whether or not anyone draws it on the architecture diagram. No common attacker has been established between the two incidents. What they share is a habit of scoping security effort by product category instead of by exposure.
What is Japan’s Government Solution Service (GSS)?
English coverage routinely calls GSS a “government cloud platform”, which flattens its structure. The Digital Agency’s own policy-evaluation material splits it into three layers. The bottom layer is a shared physical network built from dark fibre, carrier-equivalent nationwide network services, closed mobile networks and dedicated lines. Above it sit per-ministry logical networks, built on that shared foundation plus each ministry’s own building LAN and dedicated equipment. The top layer is the working environment: managed endpoints, groupware and applications.
A 2024 version of that document described the network component, GSS G-Net, as providing connectivity for 73 government organisations and municipalities, alongside a separate government-cloud connection service and an interconnection with the local-government LGWAN network. The organisations using the GSS working environment are a smaller group than those reachable through GSS networking, and mixing the two up produces bad blast-radius estimates in both directions.
How many people use GSS?
Deployment began inside the Digital Agency in fiscal 2021 and has been spreading outward since. Microsoft’s December 2025 customer study of the agency recorded about 1,500 sites and 46,400 users live at the end of November 2025, with another 2,500 sites and 150,600 users in preparation or mid-migration. The stack it describes is Microsoft 365, Teams, SharePoint, OneDrive, Power Apps, Power Automate and Copilot, supported under a Microsoft Unified contract that includes mission-critical support and named engineers.
By the time of the November disclosure, Asahi reported the service in use by about 150,000 people across 25 organisations, including the Ministry of Internal Affairs and Communications and the Cabinet Office. The Ministry of Foreign Affairs and the Ministry of Defense do not use it. Asahi also notes that the Nuclear Regulation Authority is a GSS user but keeps highly sensitive nuclear-material-protection information on a separate system, and it quotes the agency as seeing no risk that such material was in the exposed files. Growth is not finished: the September budget request earmarked funding to expand GSS to roughly 280,000 users by 2031, inside a fiscal 2027 request of about ¥1.03 trillion.
What data is GSS allowed to hold?
The confidentiality ceiling is the detail most English coverage skips. GSS is authorised to handle information up to level 2 on the government’s three-level scale. Under the Unified Standards, confidentiality-3 information must be handled as a secret document under the administrative document management guideline. Confidentiality-2 covers everything else that is likely to contain information in the non-disclosure categories of Article 5 of the Act on Access to Information Held by Administrative Organs.
Those categories include personal information that identifies individuals, corporate information whose release would harm legitimate competitive interests, information affecting national security or trust with other states, law-enforcement information, and internal or inter-agency deliberation material whose disclosure would unduly damage frank exchange of views. That last category matters here. “Not classified” and “not sensitive” are different statements, and internal meeting materials sit close to the definitional centre of confidentiality-2. A leak of conference decks is a real problem even when nothing in it is a state secret.
The September GSS data breach: how the VPN intrusion unfolded
The agency’s September notice is unusually specific about mechanics. On 25 June it detected that a large volume of files on a server had been accessed using an account belonging to a maintenance and operations person, and it opened an investigation. On 9 July it established that a third party had entered the system by exploiting a vulnerability in a network connection device, a VPN appliance, and had been accessing it without authorisation. That same day it suspended the maintenance account and cut external communication with the compromised device. The initial measures listed in the English version were the VPN patch, the account suspension and the communication cut-off. The agency reported to the Personal Information Protection Commission on 15 July, six days after confirming the intrusion, and disclosed publicly on 11 September.
📬 Stay Ahead of Cyber Threats
Get the latest cybersecurity news, critical vulnerabilities, threat intelligence, tutorials, and exclusive giveaways delivered straight to your inbox. No spam. Unsubscribe anytime.
What data was exposed in September?
About 246,000 records were exposed. Roughly 189,000 belonged to staff of GSS-using organisations and public officials involved in their work, including employees of independent administrative institutions, and roughly 57,000 belonged to businesses and individuals involved in that work. By attribute, with overlaps, the agency counted about 236,000 names, 231,000 email addresses, 94,000 telephone numbers and 1,000 addresses. My Number identifiers, financial account details and pension numbers were confirmed absent, as was personal information of ordinary members of the public.
The agency added a clarification that changes the risk picture. Most of the phone numbers and addresses were registered office locations and official contact points, because the data came from user-registration applications that ask for a representative’s name, a business phone number and a workplace address. It is still a high-quality targeting list for spear-phishing a defined population of government staff and contractors, and the agency said as much in its own warning about impersonation mail.
Why the 14-day and 64-day delays have different causes
I would not lump the delays together as “slow”. The first is the fourteen days between detecting bulk file access on 25 June and confirming it was hostile on 9 July. At his 15 September press conference, then-Digital Minister Hisashi Matsumoto explained that the 25 June signal showed only that many files had been accessed. That access could have been legitimate business, telling the two apart required fact-finding and scoping, and cutting off a legitimate account would have disrupted government work. He named early detection and analysis of unauthorised access that abuses legitimate accounts, and faster containment, as the capability the agency needed to strengthen.
The second interval is the sixty-four days from confirmation to disclosure, and his reasoning there was different. Releasing partial findings would expose the agency’s investigative capability to the attacker, so the agency waited until the investigation concluded. He acknowledged criticism of the timing and said the agency did not consider the decision wrong.
The first explanation admits a detection-engineering gap, and it is the more useful of the two. The second is a coherent counter-intelligence argument that optimises for the wrong party. By 9 July the files were already outside the perimeter, and the attacker knew what it had taken. The people who went without information were the roughly quarter-million individuals who could have tightened their own mail handling. The agency’s September advice was to avoid clicking links in unexpected messages and to expect no requests for passwords by email or phone, and that advice would have been useful on the day it was issued.
The VPN vulnerability behind the GSS intrusion: what is known and what is inference
The Digital Agency has declined to name the product or the CVE, saying that doing so would help attackers. It has confirmed that the flaw was publicly known before the attack and that its published severity at the time was Medium on CVSS. It has also confirmed something that deserves reporting: the agency moved faster than the response normally expected for that rating and was still exploited before the fix went in. Matsumoto said the same thing at the podium. The agency knew the vulnerability existed and was working through fixes in order of urgency when the attack came.
The most detailed public hypothesis comes from the Japanese security researcher piyokango, who inferred that the flaw may be CVE-2026-0257, a PAN-OS GlobalProtect authentication bypass. He gives three grounds. The published CVSS v4.0 score of 4.7 (Medium) matches the agency’s description. Exploitation was observed from mid-May, which overlaps the agency’s account that the intrusion began around late May. And the vulnerability yields an unauthorised VPN connection from outside, which fits the stated entry point of a VPN appliance used for external maintenance access. Piyokango says plainly that this is his own inference, that the agency has never disclosed the VPN product, and that no official source has mentioned this CVE. I treat it the same way. Everything below describes the vulnerability class, and none of it is a confirmed account of how GSS was entered.
How the GlobalProtect authentication override cookie flaw works
The flaw is a trust error in cookie handling, with no parsing bug or memory-safety error involved. GlobalProtect’s authentication override feature issues a cookie to an already-authenticated user so that later connections to the portal or gateway can skip credential re-entry. Rapid7’s analysis says it works like a bearer token and is not enabled by default. Exposure needs override cookies enabled and a particular certificate configuration: the vendor’s mitigation guidance says the certificate used for override cookies must not be the portal or gateway’s own certificate and must not be shared with other features.
Rapid7’s reverse engineering of the gpsvc binary explains why. A POST to /ssl-vpn/login.esp carrying portal-userauthcookie or portal-prelogonuserauthcookie routes authentication through main_AuthWithCookie, which calls main_DecryptAppAuthCookie. That function base64-decodes the cookie, decrypts it with the private key, and then trusts the recovered username, domain, host ID, client OS, remote address and timestamp, with no signature or integrity check after decryption. The advisory classifies the weakness as CWE-565, reliance on cookies without validation and integrity checking.
Because the cookie is protected by encryption with no keyed integrity check, anyone holding the public key can produce a cookie the appliance will decrypt and accept. If the override certificate is reused for the portal or gateway’s HTTPS service, any unauthenticated visitor receives that public key during the TLS handshake. Rapid7 Labs published a proof-of-concept that walks the certificate chain served on port 443 and forges an override cookie with each public key in turn until one is accepted, defaulting to the username admin.
Rapid7’s MDR telemetry shows what that looks like on the wire. It includes GlobalProtect log lines reading gateway-auth,login,Cookie,,admin from hosting-provider addresses, a consistent spoofed client MAC of aa:bb:cc:dd:ee:ff across two waves, and machine names GP-CLIENT on Linux and DESKTOP-GP01 on Windows. Where a tunnel actually came up, the logs show POSTs to /ssl-vpn/hipreport.esp and /ssl-vpn/getconfig.esp. In eight of ten affected MDR customers, the forged cookie was accepted without a full VPN session being established.
The patch compatibility switch that can undo the fix
Two operational details in the vendor guidance decide whether a patch programme works. The fix regenerates cookies using a more secure method, so every GlobalProtect user must re-authenticate once after the upgrade. In a mixed-version estate, operators can temporarily disable strict cookie HMAC validation on upgraded firewalls with set global-protect enable-auth-override-cookie-hmac no to avoid authentication failures during a phased rollout, and the guidance says to re-enable it once every portal and gateway is fixed.
That is a legitimate compatibility escape hatch, and it also silently reverts the security property the patch exists to provide. An organisation that used it during a phased upgrade and never closed the loop has patched in the asset register and not in the running configuration. The fix is a configuration state, which a version number does not capture, yet different teams track the two.
Why a Medium CVSS rating put the flaw in the wrong patch queue
The usual critique is that CVSS base scores describe a defect in isolation and say nothing about what sits behind it. For CVSS v4.0 that is only partly right, and the correction matters. The current advisory vector is AV:N/AC:L/AT:N/PR:N/UI:N/VC:L/VI:N/VA:N/SC:H/SI:H/SA:N/E:A/..., which separates impact on the vulnerable system from impact on subsequent systems. Confidentiality impact on the firewall itself is Low, while confidentiality and integrity impact on what lies beyond it are both High. The scoring system already has a way to say the appliance is not the prize.
Palo Alto published on 13 May with a 4.7 Medium rating. After exploitation was observed, the advisory carried 7.8 High with a suggested urgency of Highest. The vendor’s FAQ explains that under v4.0 the overall severity tracks the Threat score, which moves with exploit maturity, while the Base score reflects static technical characteristics. Rapid7 dates the earliest observed exploitation to 17 May, four days after publication, and recorded the CISA Known Exploited Vulnerabilities (KEV) listing on 29 May.
So the scoring standard had the vocabulary. The operational queue was keyed to the rating available on day one, and that rating assumed the exploit was unreported, an assumption reality overturned within four days. A severity-ordered queue does not re-sort itself when the threat metric changes, when a proof-of-concept ships or when a national catalogue lists the bug as exploited.
The configuration precondition makes it worse. The advisory lists Attack Requirements as None and documents the required cookie-and-certificate configuration in a separate section, which is correct scoring practice and no help to a defender without a per-device inventory of whether override cookies are enabled and whether the certificate is reused. Answering “are we exposed?” took a configuration query, and a score lookup could not answer it.
What exposure-weighted vulnerability management looks like
The agency’s own remediation commitment now reads in these terms. It says it will strengthen vulnerability management so that, given the importance of government information systems, it can respond rapidly and pre-emptively on the basis of substantive risk and not the published rating. I think that is the right correction, and it has operational consequences. An internet-facing remote-access concentrator that terminates outsourced maintenance sessions needs a different service-level objective from an internal component with the same nominal score. KEV listings, exploit-maturity changes and public proof-of-concept code should trigger re-prioritisation without waiting for the next cycle. Configuration state, not firmware version, should be the unit of compliance. None of this is exotic, but a severity queue does not do it by default.
The November GSS file-sharing leak: a different kind of failure
The November announcement describes a service used to exchange large files such as meeting materials, both between GSS users and with people outside their organisations, including contractors. Nikkei’s reporting, as summarised in piyokango’s write-up, describes the sharing function as running on a commercial cloud service, with recipients including municipalities and companies, and indicates the access happened multiple times. NHK and Nikkei reporting also indicates the investigation is checking whether highly sensitive documents were among the files. The agency told TBS it considers the likelihood of ordinary citizens’ personal information being in the exposed files to be extremely low. Yomiuri reported that the agency’s view is that GSS itself had no security problem on this occasion, a distinction that does not appear in the formal notice, which simply describes the service as provided within GSS.
Together, those reports suggest the November incident is not a rerun of September. The compromised surface appears to sit at the application and sharing layer rather than the network core, and nobody has asserted a connection between the two cases. The calendar is what bothers me. The access window opens on 30 July, twenty-one days after the September intrusion was confirmed and contained, and it runs through the investigation, the victim-identification work and the 11 September disclosure. Whether or not the two are causally related, an organisation can be running the most intense security review in its history while a different service on the same platform quietly serves files to the wrong people for three and a half months.
Why file-sharing platforms are hard to defend
The difficulty comes from the service’s specification, and it is no defect. A VPN gateway is supposed to admit only authenticated, authorised principals, so anomalous authentication is a meaningful signal. A file-sharing service exists to let a contractor, municipal officer or vendor retrieve an object, often through a link, a guest identity or a federated arrangement, and often from a network the operator has never seen. Reachability is the feature.
So the controls that matter sit outside the perimeter. They are the scope and lifetime of each share, whether access is bound to a verified identity or to possession of a URL, whether authorisation is re-evaluated on every object request, whether uploaded content is classified before it can be shared, and whether per-object read events reach an analyst with enough context to tell a contractor’s first download from an unindexed crawler walking a link tree.
Japan’s own baseline already states the elementary version. The Unified Standards’ section on using cloud services to share information with people outside the organisation gives two countermeasure examples: set the sharing scope to only those who need it, and promptly delete information that is no longer needed. Those are the right two controls, written as procedural expectations and not as platform-enforced invariants. A rule that lives in a policy document depends on the diligence of individual officials under deadline pressure, one share at a time, across 150,000 users. A rule that lives in the platform is enforced by default-deny external sharing, mandatory expiry, identity-bound links and automatic purge. To me, the 105-day window is what that difference looks like from the outside.
Zero trust on paper versus zero trust in operation
Asked how this could happen on a zero-trust platform, Matsumoto answered that however hard you search, a hole can exist somewhere, which is why zero trust requires continually revisiting preventive measures. That is honest, and it is incomplete, because it treats zero trust as a state you reach. It is better understood as a property of where enforcement points sit and what they inspect.
The agency’s own 2024 description of GSS security is candid about the design. It adds zero-trust endpoint defence on top of conventional boundary defence, summarised as “boundary defence plus endpoint defence”, with continuous monitoring of device and communication behaviour and immediate cut-off to prevent lateral movement. In response to the September questions, the agency declined to describe GSS’s actual security configuration or system architecture, again on the grounds that disclosure would benefit attackers.
What Japan’s Unified Standards actually require
The most revealing document here is a standard, not a press release. Part 7.3 of the Unified Standards defines the zero-trust element as dynamic access control: for a given access, evaluating and verifying the requester’s trust information afresh for each operation that has no established session. It then lays out an implementation method. Identify the target systems, identify their resources, enumerate access patterns, risk-assess those patterns, and select the patterns on which to implement dynamic control. It also states that an organisation that does not implement dynamic access control need not comply with 7.3.1 and 7.3.2.
So the national baseline allows selective adoption, scoped by a risk assessment the organisation performs itself. That is a defensible position for a baseline covering hundreds of heterogeneous bodies. It also means the phrase “zero-trust architecture” in a procurement document guarantees nothing about any particular request path.
Elsewhere the same standard is quite prescriptive about the controls the September incident needed. It calls for multi-factor authentication on remote access and administrator logins, least-privilege assignment, dual-authorisation mechanisms for completing sensitive operations, and a ban on using administrator privileges for ordinary business. It also calls for automated log aggregation, time-series presentation and reporting, and monitoring aimed at lateral movement inside the environment.
A maintenance identity doing bulk reads across a file server is the canonical case those clauses exist to catch. Controls present on paper and absent in effect describe this incident well. A maintenance account is by construction an exception to ordinary authorisation policy. If that exception is standing instead of just-in-time, broad instead of scoped, and unrecorded instead of session-captured, it becomes a trust anchor the architecture never meant to create. It is reachable through whichever appliance terminates maintenance connectivity, which means it is reachable through the patch queue.
Detection gaps: what would have caught each GSS incident sooner
Both incidents were eventually detected, so the fair framing is telemetry that did not turn into a decision quickly enough, and not a lack of telemetry. In September the signal existed on 25 June and stayed ambiguous for fourteen days because the organisation could not immediately say whether a maintenance identity reading many files was doing its job. That ambiguity is a design property and no bad luck. It is what happens when human operators and service processes share account semantics, when per-identity volume baselines are not pre-computed, and when no ground truth exists, such as canary objects, decoy shares or known-bad read patterns, that would fire an unambiguous alarm on the first hostile read and not the thousandth. The minister’s own remediation language, about early detection and analysis of unauthorised access abusing legitimate accounts, describes this gap.
For November, the relevant signals are object-level and recipient-aware, not host-level. An analyst needs to know how many distinct objects a given recipient identity or link has read in a window, whether the link is being used from networks and geographies that have never used it before, whether the ratio of downloads to previews has shifted, whether dormant shares have suddenly become active, and whether requests are arriving for objects whose intended recipient list excludes the requester.
None of this needs exotic machine learning. It needs per-object audit events with recipient context, retained long enough to baseline, and thresholds tuned to identity behaviour and not to aggregate traffic volume. The 105-day window bounds what the deployed detection was able to see, whatever it turned out to be, and the agency says the detailed chronology of that period is still under investigation.
The week around the announcement: Microsoft 365 outage, Studio and Japan’s policy response
The November disclosure landed in a crowded period, and the surrounding events sharpen the structural argument. On 22 October the Digital Agency had already disclosed that Studio, the commercial website-building service used for part of its own web presence, suffered unauthorised access to its cloud environment, with up to 13,323 records from inquiry forms potentially exposed.
On 13 November a global Microsoft 365 outage that originated in Exchange Online took down more than twenty services. GSS mail became temporarily unavailable, and the Prime Minister instructed ministries to secure non-email contact channels, protect citizens from missing application deadlines and report sector impacts. At his 17 November press conference, Digital Minister Toshiharu Furukawa confirmed the GSS mail disruption and argued for keeping multiple external communication channels as a standing capability, not an emergency improvisation. The same week brought the disclosure of unauthorised access to RIKEN’s Fugaku supercomputer, and Furukawa was questioned on that in the same session.
Japan’s Cybersecurity Emergency Response Package
The policy response has run in parallel. On 6 November the Prime Minister convened the second ministerial meeting on cybersecurity and announced a Cybersecurity Emergency Response Package built on four pillars: thorough basic measures using AI, stronger intelligence collection and investigative and analytical capacity, strategic execution of cyber response, and stronger victim support and damage limitation. Roughly ¥10 billion in contingency reserves backs it.
Instructions to Furukawa included examining revision of the Cyber Response Capability Enhancement Act, shortening the time for critical-infrastructure operators’ reports to reach the National Cybersecurity Office to within hours and at the latest the same day, and developing mechanisms for analysing suspicious communications at private companies. A separate instruction directed a review of government systems for IT supply-chain risk, led by the National Security Secretariat. The oversight body is recent: the NCO was established on 1 July 2025 by reorganising NISC after the Cyber Response Capability Enhancement Act was enacted in May 2025.
An unresolved governance question
One governance wrinkle came up directly at the September press conference. Matsumoto held both the digital portfolio and the cyber-security portfolio. When asked who would independently review the agency’s handling of the incident, he acknowledged the objection that he supervised both the reviewed and the reviewer, while arguing that the NCO had worked in a substantially independent manner. Furukawa holds both roles today.
The handover itself happened mid-crisis. Matsumoto left the post in the 17 September reshuffle, and Furukawa, a surgeon and lawyer entering cabinet for the first time, replaced him. No architecture diagram addresses continuity of technical ownership across a political handover in the middle of an active investigation.
Concentration risk in Japan’s government cloud
Taken together, the pattern I see is concentration, and I do not read it as incompetence. Consolidating twenty-odd ministries onto one network, one identity fabric, one productivity suite and one sharing service bought uniformity, purchasing leverage and a genuinely better baseline than the ministry-by-ministry estate it replaced. It also made failures correlated. The November outage showed that for availability: one provider’s health-check regression removed email from a large part of the Japanese state at once. The two disclosures show the same property for confidentiality.
I do not think this makes consolidation a mistake. It moves risk from many small, uncoordinated estates into a few large, coordinated ones, and the security organisation has to be resourced and structured for that second regime. The Digital Agency employed about 1,220 people as of July 2026, roughly 580 career administrators and 640 hired from the private sector. The open question is how much of that capacity goes to running enforcement points and reading their output, and how much goes to building the next migration.
What would change the outcome for government cloud security
The commitments that would matter are narrow and testable.
Every externally reachable interface needs to be its own trust boundary. The remote-access concentrator, the maintenance plane behind it, the sharing endpoints and the guest identity path each need their own exposure-weighted remediation clock, least-privilege model and detection thresholds. One platform-wide policy applied uniformly is too slow for the edge and too coarse for the sharing layer.
Maintenance access should become a brokered, time-bounded, dual-authorised, session-recorded exception in place of a standing credential. Then the compromise of an operator yields a narrow window and no bulk-read capability, and the fourteen-day ambiguity of 25 June cannot recur.
Sharing controls need to move from the policy document into the platform: default-deny external sharing, mandatory expiry, links bound to verified identities in place of URL possession, prompt automatic deletion, and per-object audit events flowing to an analyst with recipient context. The Unified Standards already asks, in plain language, for limiting the sharing scope and deleting information promptly. The gap is in enforcement.
Four metrics would show whether any of this took hold: elapsed time from vendor advisory to verified configuration state on internet-facing edge devices, re-prioritisation latency when a flaw’s exploit maturity changes, the proportion of externally shared objects that carry both an expiry and an identity binding, and elapsed time from the first anomalous object read to containment. September’s numbers were fourteen days to classify and seventy-eight to disclose. November’s were 105 days of access and eight days to disclose. The disclosure clock improved markedly, while the containment clock is the one that decides how much data leaves.
Neither incident required a zero-day, a novel technique or an implausibly sophisticated adversary. If piyokango’s inference is right, September used a publicly documented authentication bypass, whose proof-of-concept iterates a TLS certificate chain, against a device whose exposure depended on a configuration choice, and then a legitimate account performing its normal kind of operation at an unusual volume. November involved a service designed to give files to outsiders. The controls that would have mattered are known, written down in Japan’s own baseline, and in several cases explicitly optional. The gaps sit at the externally reachable edge, the maintenance plane and the sharing layer, which is where the daily work happens.
Frequently asked questions about the Japan GSS data breaches
What happened in the November 2026 GSS file-sharing leak?
On 20 November 2026, Japan’s Digital Agency said files in the file-sharing service it provides through GSS may have been accessed by people other than the intended recipients. It detected the access on 12 November and suspended the service the same day. The access ran from 30 July to 12 November, a window of 105 days. The agency has not disclosed the number of files, their contents, the affected ministries or the intrusion path.
How many records were exposed in the September 2026 GSS breach?
About 246,000 records. Roughly 189,000 belonged to staff of GSS-using organisations and public officials, and roughly 57,000 belonged to businesses and individuals involved in that work. The agency counted about 236,000 names and 231,000 email addresses. My Number identifiers, financial account details, pension numbers and the personal information of ordinary members of the public were confirmed absent.
Which vulnerability was exploited in the September GSS intrusion?
The Digital Agency has not named the product or the CVE. It has said the flaw was publicly known, rated Medium on CVSS at the time and exploited in a VPN appliance before the fix was applied. Researcher piyokango has inferred it may be CVE-2026-0257, a PAN-OS GlobalProtect authentication bypass, and says no official source has confirmed this.
What level of confidential information can GSS handle?
GSS is authorised to handle information up to confidentiality level 2 on Japan’s three-level scale. Level 3 covers information handled as a secret document, and GSS is not authorised for it. Level 2 still includes personal information, corporate information, law-enforcement information and internal deliberation material that falls under the non-disclosure categories of Article 5 of the Act on Access to Information Held by Administrative Organs.







