ThreatBeaconXSubscribe
HighAPT · 15 min read

From Firewall to Cloud: Tracing the k0t Attack Chain

A joint investigation by Group-IB and NyxLab links the k0t activity cluster to exposed firewalls, VPN weaknesses, stolen credentials, unauthorized remote-access tools and cloud-based data theft. This article explains the different entry methods, separates confirmed findings from assumptions, and shows why recovery must cover identities and internal systems—not just the device first compromised.

Written by Yedla Sai Geethika·Published Oct 25, 2026

Group-IB and NyxLab published their investigation of the k0t activity cluster on October 21, 2026. The report describes an operation that finds exposed security appliances, obtains access, collects credentials and expands into internal networks.

Researchers identified more than 70,000 exposed credentials and related records. This is not a count of 70,000 breached organizations. The investigation also found remote-access infrastructure, credential-theft tools and evidence of data being transferred outside some compromised environments.

The central issue is not one malware file or one product vulnerability. It is a sequence in which a security appliance becomes a source of access to other systems:

Find an exposed device → Obtain access → Collect credentials → Enter the internal network → Establish remote control → Steal data.

What “Network Edge” Means

The network edge is the boundary between an organization’s internal systems and the internet. Firewalls and remote-access VPN services often operate at this boundary.

A VPN gives an authorized user a protected connection into a private network. A management interface lets an administrator change a device’s settings.

These are different functions. Employees may need an internet-accessible VPN portal, but the device’s administration page does not necessarily need the same exposure.

The products mentioned in this investigation also have different purposes:

ProductSimple explanation
FortiGateA network-security appliance that can provide firewall and VPN functions.
FortiWebA web application firewall designed to protect websites and web applications.
GlobalProtectPalo Alto Networks’ remote-access technology for connecting users to organizational resources.

They must not be treated as one product with one shared vulnerability. A flaw affecting FortiWeb does not automatically affect FortiGate.

The Findings—and the Limits of the Numbers

Reported observationCorrect interpretation
More than 70,000 exposed credentials and related dataRecords associated with the investigation, not a verified count of unique people or companies.
84 devices in 28 active MeshCentral groupsA snapshot of one management environment examined by researchers.
Evidence of 100 removed device groupsHistorical group records, not proof of 100 additional distinct victims.
37.6% of identified unique domains associated with Asia-PacificA dataset distribution, not proof of deliberate regional targeting.
Overlap with Qilin and The Gentlemen leak-site victimsA reason to investigate possible relationships, not proof that k0t is either ransomware group.

The source also places many FortiGate compromises between February and June 2026. Its October publication date should not be mistaken for the start of the activity.

Why Many Different Organizations Could Be Targeted

The reported reconnaissance used internet-exposure search services, including FOFA and ZoomEye, rather than depending entirely on research into a particular company. Other product searches showed interest, but interest alone does not establish successful compromise. 

The practical meaning is straightforward: an attacker does not need to know your business before finding a reachable login page or vulnerable device.

An exposed system is not automatically compromised. However, an organization should know which services are visible, why they are visible, who owns them and how they are maintained.

MITRE recognizes public scanning databases as a reconnaissance source because they can reveal addresses, open services, certificates and other information useful for selecting targets. 

Additional defensive analysis: A small business should not assume it is safe because it is not a famous target. In an exposure-led campaign, being reachable with a useful weakness may matter more than company size.

Three Entry Problems, Not One

Entry route discussed in the reportPlain-language explanationImportant boundary
FortiWeb: CVE-2025-64446A flaw can allow administrative actions without a normal authenticated login.This CVE affects FortiWeb, not FortiGate.
FortiGate credential attacksAttackers try to obtain access through weak or reused credentials.A successful login is not necessarily proof of vulnerability exploitation.
GlobalProtect: CVE-2026-0257Vulnerable authentication-cookie handling can permit unauthorized VPN access.Exposure depends on the PAN-OS version and relevant configuration.

The campaign source distinguishes these routes. Fortinet and Palo Alto Networks provide separate advisories and remediation requirements. 

FortiWeb: The Risk of Unauthorized Administration

Fortinet describes CVE-2025-64446 as a path-traversal weakness that can permit unauthenticated administrative commands.

In simple terms, the service can mishandle a request in a way that reaches a function the requester should not control.

The relevant defensive question is not only whether the update is installed. Administrators also need to look for changes made before patching, including unfamiliar administrator accounts and altered access restrictions.

Fortinet lists these minimum fixed releases for this specific vulnerability:

FortiWeb branchMinimum fixed release
8.08.0.2
7.67.6.5
7.47.4.10
7.27.2.12
7.07.0.12

These are vulnerability-specific minimums, not a recommendation to install an old build today. Use a supported, vendor-recommended release and its approved upgrade path.

Fortinet also advises restricting exposed management access and reviewing configurations after upgrading. 

FortiGate: Why Updates Do Not Repair Password Reuse

Fortinet’s separate June 19, 2026 assessment of activity called FortiBleed emphasized credential reuse and brute force rather than a new Fortinet vulnerability. Its guidance includes ending active sessions, resetting exposed credentials, applying MFA and checking unauthorized configuration changes. 

This explains why “we installed the latest firmware” is an incomplete answer to a credential-based incident.

A patch repairs a software defect. It does not necessarily invalidate a password already known to an attacker, remove an unauthorized account or recover a configuration file copied earlier.

Repeated failures, temporary lockouts and later successful access should therefore be reviewed together. A lockout is a protective control, but it should also create an investigation signal—not simply a waiting period after which monitoring starts from zero.

GlobalProtect: A Login Cookie Can Become an Access Credential

An authentication cookie is information a service uses to remember an earlier login. A flaw in how that proof is validated can let an attacker obtain access without completing the expected login process.

Palo Alto Networks says CVE-2026-0257 affects particular GlobalProtect portal or gateway configurations using authentication-override cookies and a specific certificate arrangement. It is not a claim that every GlobalProtect deployment is vulnerable. 

The vendor advises installing the appropriate fixed PAN-OS release. Temporary options include disabling authentication override or using a newly generated certificate dedicated exclusively to that purpose.

Participating portals and gateways need a coordinated update, and users may need to authenticate again afterward.

Additional defensive analysis: MFA remains valuable, but it cannot substitute for fixing an authentication bypass. A service that incorrectly accepts a session credential may never reach the normal MFA challenge.

Why a Stolen Configuration Can Matter More Than the Firewall

A device configuration is more than a list of settings. It can reveal network structure, service accounts and other information useful for further access.

MITRE explicitly documents network-device configuration collection as a technique for obtaining such information. 

A particularly important account is an LDAP or Active Directory bind account. The appliance may use it to communicate with the organization’s directory—the system that keeps track of users, computers and access rights.

A useful analogy is that the firewall has its own staff badge for checking the company directory. If that badge is exposed, repairing the firewall does not automatically cancel the badge.

This is why incident response should identify each credential referenced in an exposed configuration, determine its permissions and review where else it was used.

Fortinet specifically recommends treating AD/LDAP integration accounts as compromised when appliance compromise is identified. 

A Necessary Clarification About Configuration Encryption

The Group-IB report describes offline recovery of secrets from copied configurations. Fortinet’s underlying advisory, FG-IR-19-007, is narrower than the statement “all firewall passwords can be decrypted.” 

CVE-2019-6693 concerns a hard-coded encryption key protecting affected FortiOS ENC credential fields.

The advisory explicitly excludes the administrator password from that particular FortiOS issue. Administrator credentials can still be compromised through other methods; this specific weakness should not be used to claim that every administrator password is reversibly encrypted. 

Fortinet introduced configurable private-data encryption and recommends protecting configuration backups. The correct implementation depends on the product version and environment; high-availability and centrally managed deployments require compatibility checks.

Practical implication: Enabling stronger protection now does not repair an older copy already held by an attacker. Review historical exposure and rotate the secrets that copy could reveal.

Keep recovery keys protected separately and test restoration before changing production encryption settings.

What Happens After the VPN Connection

OpenConnect is a legitimate VPN client that supports GlobalProtect and FortiGate protocols. Its presence alone is not evidence of malware. 

The concern is an unauthorized person using a valid-looking network connection. Once inside, attackers may find systems that trust the VPN address range or give service accounts broader access than necessary.

The k0t report describes internal discovery, credential theft, privilege escalation and remote execution using established administration and security-testing utilities. 

Tool named in the sourceRole described in the investigation
NetExecInternal discovery, credential access and remote execution.
SoftPerfectIdentifying hosts, services and shared resources.
ImpacketCredential-related and Windows/Active Directory operations.
KerbruteDomain-account password attacks.
noPacPrivilege escalation involving CVE-2021-42278 and CVE-2021-42287.
PetitPotamCoercing authentication as part of a relay-related chain.
MimikatzAccessing authentication material in Windows memory.

These names explain reported behavior. They are not a claim that every tool appeared on every victim. Legitimate testing teams also use several of them.

Understanding the Windows Terms

Active Directory is a centralized system for managing users, computers and permissions. A domain controller is a server performing that role. Domain-administrator privileges can provide extensive control across the environment.

The source refers to several credential stores and mechanisms:

TermSimple meaning
SAMA Windows database containing local-account credential information.
LSASSA Windows security process whose memory can contain authentication material.
LSA secretsStored operating-system or service secrets.
DPAPIA Windows mechanism used to protect application data.

MITRE distinguishes these credential-access techniques. 

The useful point is not to memorize every acronym. It is to recognize the escalation:

One appliance account → Other credentials → Broader permissions → Access to more systems.

A low-privilege directory account limits potential damage. A firewall should not be given domain-administrator rights merely because that makes integration convenient.

Remote-Management Software Becomes a Second Entrance

MeshCentral provides legitimate remote desktop, terminal and file-management capabilities. An installed MeshAgent connects an endpoint to a management server. 

In this campaign, the researchers found unauthorized MeshCentral access alongside AdaptixC2 and other remote-management or tunneling tools. Agent downloads and management infrastructure were separated. 

The distinction matters: blocking the original download site does not necessarily stop an agent already installed and communicating elsewhere.

For each remote-access installation, defenders should establish its approved owner, installation source, management destination, account membership and business purpose.

MITRE recommends reviewing the behavior chain, including execution, persistence, sustained outbound communication and interactive child processes. 

The software name is not the security decision. Authorization and behavior are.

What the MeshCentral Snapshot Does—and Does Not—Prove

The reported 84 devices and 28 active groups demonstrate that researchers observed an organized remote-access environment. They should not be used to calculate a final number of affected companies.

Removed groups also cannot simply be added to current groups and called verified victims. 

The source interpreted the lack of server-side stored files as supporting an interactive-access role, while separately observing Rclone exfiltration.

Analytical limitation: Absence of stored files alone cannot prove that no live file transfer ever passed through a remote-management service.

That distinction matters during incident response. Review endpoint and network records rather than using one empty folder on a control server to conclude that data was safe.

Why a Separate C2 Channel Increases the Recovery Work

A command-and-control server, or C2, lets an operator send instructions to a compromised machine.

The report identifies AdaptixC2 as another channel and describes a fake installer named:

clear-thinking_61.92.4_INSTALL.exe

as part of agent delivery. 

Additional defensive analysis: When an intrusion has more than one remote-control route, removing the first detected agent is not proof of eviction. Investigators need to map all access methods, associated accounts and affected hosts.

Generic paths such as /update, or a common self-signed certificate name, are weak signals by themselves. Combine them with the specific destination, process, installation history and other evidence.

A common path should not trigger a declaration of compromise on its own.

Cloud Storage Can Be Used to Remove Data

The investigation reports Rclone transfers to Wasabi S3 storage and MEGA in some compromised environments.

Rclone itself is a legitimate file-management utility supporting many cloud services. It is not malicious simply because it is installed. 

The question is whether the transfer was authorized: which process initiated it, which account owned the destination, which files were read, and whether the activity matched an approved backup or migration job.

MITRE distinguishes exfiltration to cloud storage from communication through the primary C2 channel. Defenders should therefore not look only at the remote-control server when investigating possible data loss. 

A useful correlation is:

Unusual VPN access → Broad file access → Unexpected transfer utility → Upload to an unapproved cloud account.

This is a proposed hunting sequence, not a tested rule with a published detection rate.

The Ransomware Connection Needs Careful Wording

An initial access broker obtains access to organizations and may offer that access to other criminals.

The k0t report considers a broker role plausible because of the operation’s artifacts and victim overlap with Qilin and The Gentlemen. Direct affiliation or a specific access sale remains unconfirmed.

The safe conclusion is not “k0t is Qilin.” Nor is every device seen in the report a confirmed ransomware-encryption victim.

The operational concern is that usable access can outlive its first operator. An organization should investigate credential theft and persistent access before a ransom note appears.

How This Relates to FortiBleed

Fortinet’s June 2026 FortiBleed assessment focused on reused credentials and brute force, not a newly disclosed Fortinet flaw.

Group-IB and NyxLab compared their data with the previously reported dataset and found negligible overlap. They assessed a possible relationship to broader FortiBleed-related activity, but did not establish common ownership of the datasets. 

Shared vendors, similar timing and similar techniques are useful leads. They are not enough, by themselves, to merge two investigations into one actor.

Related Research: Review More Than User Workstations

A separate Unit 42 report, updated October 20, 2026, described other intrusions involving CVE-2026-0257. It discussed movement toward virtualization infrastructure and unintended management exposure.

Those observations are wider context—not confirmed additional steps in the k0t campaign.

The defensive implication is to inspect the actual access available from a VPN session.

Can it reach directory servers, backup systems, hypervisor management or other high-value services? Do not assume workstation endpoint protection covers every system reachable from the VPN.

Segmentation should limit that reach before an incident and help define the investigation scope after one.

Additional Analysis: Five Conditions for Closing the Incident

The original report already emphasizes enterprise credential exposure and correlated monitoring. The framework below organizes those lessons into recovery checks; it is not a claim of a new attack discovery.

Recovery areaQuestion that needs an evidence-based answer
ExposureIs the original vulnerable or unnecessarily exposed interface remediated?
IdentityWere exposed accounts, tokens and sessions reviewed and replaced or revoked as needed?
PersistenceWere unauthorized agents, accounts and other continuing access methods removed?
DataWas the scope of unauthorized access and transfer investigated?
VerificationWas the restored environment rechecked through independent logs, configuration review and endpoint evidence?

A hypothetical example shows why this matters.

Suppose a firewall is patched, but an exposed directory account remains valid and an unauthorized agent remains on a server. The original weakness is fixed, yet the attacker may still have other access.

Fixing the entrance is necessary. Proving that the intruder has no remaining access is a separate job.

Detection Priorities for Security Teams

Evidence sourceWhat to reviewWhy context matters
Appliance administrationNew administrators, trusted-host changes, configuration exports and access from unusual sources.A legitimate backup or approved change can look similar.
VPN and identityUnexpected sessions, new client patterns, service-account use and privilege changes.A successful login does not prove the legitimate owner initiated it.
Windows and directory serversUnusual credential-store access, remote tasks/services and broad discovery.Approved administration and security testing must be distinguished.
Remote-access toolsInstallation source, management server, owner and persistent connections.The same software may be authorized on one host and unauthorized on another.
Cloud transfersProcess, file-access history, volume and destination account ownership.A reputable provider can host an attacker-controlled account.

This combines the campaign’s hunting recommendations with MITRE guidance for configuration collection, remote-access abuse and cloud exfiltration.

fort.py
fortnew.py
exploit_2026_0257.py
clear-thinking_61.92.4_INSTALL.exe
/update
/api/v1/check
/uri.php

The first three are reported attacker-side tooling artifacts, not files guaranteed to exist on a victim. The paths are generic hunting patterns, not proof of compromise by themselves.

Practical Response and Prevention

First, establish exposure.
Inventory the exact product, version, administrative interface and relevant VPN settings. Restrict management access where possible. Confirm actual exposure rather than assuming the architecture diagram is current. 

Second, preserve and correlate evidence.
Retain appliance, authentication, endpoint, DNS, proxy and cloud records. Put events on one timeline. Keep actual passwords and sensitive configuration backups out of ordinary tickets or public submissions. 

Third, repair the entry path.
Apply the appropriate supported firmware and configuration fixes. For GlobalProtect, coordinate cookie-related changes across participating components. Do not treat a generic firmware number as suitable for every product branch. 

Fourth, recover identities.
Review administrator and VPN accounts, directory bind accounts and other exposed secrets. Revoke relevant sessions, rotate credentials through a clean administrative path and verify dependent services afterward. MFA is important, but compromised session mechanisms also need remediation. 

Fifth, remove continuing access.
Validate remote-management agents and their controlling servers. Preserve evidence before removal, inspect related persistence, and rebuild systems when their integrity cannot be established. MITRE recommends restricting unapproved remote-access functionality and correlating persistent connections with interactive activity. 

Finally, verify the result.
Compare configurations with an approved baseline, confirm unauthorized access methods are gone and continue looking for use of exposed accounts. Keep a record of which systems were checked and which uncertainties remain.

What This Means for Different Readers

Employees: This campaign does not require every victim to click a phishing link. Report unexpected remote-control activity, unapproved support software and unusual VPN events. Do not investigate by running attacker tools.

Small-business owners: Ask the IT provider who can administer the firewall, whether its management page is public, how updates are verified and which remote-support tools are approved. Request evidence, not only “the firewall is working.”

Network administrators: Treat configuration backups as sensitive. Confirm both patch status and identity exposure, and coordinate encryption changes with restoration and high-availability requirements.

Security analysts: Join the appliance, VPN, identity, endpoint and transfer timeline. An event that looks ordinary in isolation may matter when it follows an unauthorized configuration export. 

What the Report Does Not Prove

The source does not establish a confirmed total of unique victim organizations, deployment of every listed tool in every environment, or direct ownership of Qilin or The Gentlemen by k0t.

It also does not confirm that an AI testing tool found among the artifacts was used against victims.

Aliases and developer labels are investigative identifiers, not established real-world identities. The source mentions k0t, CryptoFeel and Scrooge/ScroogeMcDuck with differing levels of uncertainty. Tool developers were not directly established as campaign operators.

No new discovery should be claimed merely because the report is rewritten. The useful contribution here is the separation of product-specific risks and a practical recovery model.

Threat Assessment and Conclusion

High severity is an appropriate editorial rating for this campaign. Confirmed privileged access, domain compromise or data loss may require critical internal incident handling. This rating is not a CVSS score for every vulnerability mentioned.

The most important lesson is that a network-security appliance is also a store of trust: it controls access and may hold credentials used by other systems.

Recovery must therefore answer more than “Is the firewall patched?”

It must also answer:

“Which secrets were exposed, which systems were reached, what access remains and what data left?”

Protect the device, protect the identities connected to it, and verify the whole intrusion path.

Indicators of Compromise

TypeValueNotes
IPv462[.]60[.]131[.]43—
IPv462[.]60[.]135[.]64—
IPv4193[.]149[.]187[.]81—
IPv464[.]94[.]85[.]14—
IPv4168[.]100[.]11[.]99—
IPv4162[.]33[.]178[.]213—
IPv4162[.]33[.]179[.]249—
IPv4193[.]24[.]123[.]162—
IPv4193[.]149[.]176[.]50—
Domainmesh[.]buttruyen[.]com—
Domainmeshcentral[.]girydsfsggvb[.]com—
Domainagent[.]afahimsa[.]top—
Domainsthyjk[.]com—
Domainalknene[.]shop—
Domainradesashar[.]com—
URLhxxps://agent[.]afahimsa[.]top/meshagents?id=4&meshid=DSZSDHzU1YSP83eNnHzSUfPPt5MFUC3UadztK7UJWbWttVfTo$auq8StiVu0p6ND&installflags=0—
URLhxxp://193[.]149[.]187[.]81:1287/1.exe—
URLhxxps://sthyjk[.]com:8443/clear-thinking_61.92.4_INSTALL.exe—
URLhxxps://sthyjk[.]com:8443/clear-thinking_61.92.4_INSTALL.exe—
URLhxxps://64[.]94[.]85[.]14:4321/endpoint—
SHA-256158215c7371e868f3d0c7fd2e92f05026a8678a3635629d89bd6042ee543278a—
SHA-2561d1b76d30a69aa39b224053481483156f82dde026edd6f5d964f6e03f5f04a6d—
SHA-2564db6bea4c1e4bce5020039def4087a073f4753a6b297b2832c19c10423ac842d—
SHA-256552d0096f60c36a35c6096b3e4b38d743c29f8c56f3862fa3b7ef06c932e2a21—
SHA-256f9e95af4cda14dd5596c452a1a698f60412a7b2a9540321ee2a10b3bcbbd2a0e—
SHA-25639c4decc4881ec02a47cd01c0fd8c3030527b466124ad03c6f99cfaa0ba0b72e—
SHA-256808145ec3351569083fda27a910ebdf9e054e8ca42e40c021af5d6712c2ab87c—
MD55318f29ef434341d989d543b14fd2a48—
MD520dc8f1b8a313336c14e03bcb71ff6c8—
MD5032473db4d66748eab524effe57a661d—
MD513367fbba4344de87de418ab17de76f0—
MD5197f2621c80da7974c56d2ff1921e994—
MD56f5d18157425aa9697913ae1310f588f—
MD50dc1af7cfd336a7f075228defc730082—

MITRE ATT&CK Mapping

T1596.005T1583.001T1588.005T1583.004T1588.002T1583.006T1190T1078.003T1059.001T1053.005T1569.002T1133T1136.001T1068T1550.004T1550.002T1003.006T1558.003T1003.004T1003.001T1110.003T1003.002T1187T1069.002T1083T1046T1135T1018T1518.001T1602.002T1219T1567.002