Back to Intelligence

AdaptixC2 + k0t Edge Campaign, 29-Bank Phishing Kit, and LLMjacking: OTX Credential Theft Analysis — Enterprise Detection Pack

SA
Security Arsenal Team
November 21, 2026
11 min read

Threat Summary

Three concurrent OTX pulses paint a coherent picture of a credential-theft economy operating at industrial scale in Q4 2026. The campaigns are distinct in tooling and operator, but they converge on the same commodity: authenticated access — whether to network edge appliances, consumer banking portals, or cloud billing accounts.

  1. k0t (edge exploitation campaign) — Group-IB and NyxLab attribute a large-scale operation to threat actor k0t, who is systematically compromising FortiGate, FortiWeb, and Palo Alto GlobalProtect edge devices. The actor exploits CVE-2025-64446 (FortiWeb) and CVE-2026-0257 (GlobalProtect), supplemented by custom brute-forcing tooling. More than 70,000 exposed credentials have been recovered from this operation, and post-exploitation leverages AdaptixC2 beacons alongside MeshCentral RMM infrastructure (meshcentral.girydsfsggvb.com) for persistence. The tag set — credential harvesting, initial access broker — indicates k0t is likely monetizing access rather than conducting direct espionage: harvested edge-device credentials and VPN sessions are classic IAB inventory sold on dark web forums.

  2. 29-bank phishing kit — A single phishing template kit has spawned 29 unique fake-bank domains targeting US and Canadian financial customers since late September 2026. Every cloned site shares identical artifacts: the same tagline, routing number, and support phone number — an operational fingerprint that makes clustering trivial for defenders but also proves the kit is being redeployed across disposable domains (.online, .live, .cc, .com) at velocity. Objective: credential and PII theft for account takeover and downstream fraud.

  3. LLMjacking via GCP phishing — Wiz reports threat actors purchasing sponsored search ads to push fake Google login pages, compromising privileged GCP users, then disabling organizational policy guardrails, creating service accounts with Gemini API access, and inviting external attacker-controlled accounts. The endgame is cloud billing hijacking: victims pay for attacker LLM consumption (LLMjacking), and harvested cloud credentials seed further cross-tenant exploitation.

Collective assessment: All three campaigns feed the same dark web supply chain — credentials harvested at the edge, at the consumer layer, and at the cloud control plane. Organizations should assume that any credential exposure in one layer will be tested against the others within hours.

Threat Actor / Malware Profile

k0t (Initial Access Broker)

  • Distribution / Initial Access: Exploitation of internet-facing edge appliances — CVE-2025-64446 (FortiWeb), CVE-2026-0257 (Palo Alto GlobalProtect), plus custom brute-force tooling against FortiGate administrative interfaces. Legacy CVEs in the pulse (CVE-2019-6693, CVE-2021-42278, CVE-2021-42287) indicate the actor also chains older FortiOS credential-disclosure and AD privilege-escalation flaws (sAMAccountName spoofing / noPAC) when pivoting internally.
  • Payload behavior: Deployment of AdaptixC2 beacons — an open-source post-exploitation framework increasingly adopted by criminal actors as a Cobalt Strike substitute. Beacons support shellcode injection, token manipulation, and SOCKS tunneling for lateral movement.
  • C2 communication: AdaptixC2 defaults to HTTP/HTTPS beaconing with malleable profiles; the pulse identifies MeshCentral RMM (meshcentral.girydsfsggvb.com) as a redundant access channel — a dual-C2 strategy common among IABs to survive takedowns.
  • Persistence: RMM agent installation (MeshCentral registers a Windows service), rogue local admin accounts on edge devices, and legitimate VPN credential reuse — persistence that survives patching because it rides on valid accounts.
  • Anti-analysis: Use of legitimate RMM tooling blends into enterprise admin traffic; exploitation of edge devices leaves minimal host-based telemetry since most appliances lack EDR coverage.

Phishing Kit (29-bank operation)

  • Distribution: Malvertising and lure emails; disposable domains registered in bulk.
  • Payload behavior: Static credential-harvesting clones with identical backend artifacts (shared routing number, tagline, phone number), exfiltrating credentials/PII to kit-controlled endpoints.
  • Persistence/anti-analysis: None on-host — the kit's resilience comes from rapid domain rotation, not endpoint sophistication.

LLMjacking Operators

  • Distribution: Sponsored search ads impersonating Google/GCP login portals (typosquats: groosl.com, gboogl.com, gcoalg.com).
  • Payload behavior: AiTM-style credential capture, then cloud-native abuse: org-policy manipulation, service account creation with aiplatform/generativelanguage scope, external account invitation.
  • Persistence: Service account keys and federated external identities that survive password resets unless explicitly revoked.

IOC Analysis

The pulses contain four operational indicator classes:

TypeExamplesOperationalization
Domains / hostnamesmeshcentral.girydsfsggvb.com, brayfinance.online, bank.firstcp.live, groosl.com, gboogl.comBlock at DNS sinkhole, secure web gateway, and email gateway. The GCP typosquats are high-fidelity — minimal false-positive risk.
File hashes (SHA256/MD5)39c4decc...b72e, 6f5d18157425aa9697913ae1310f588fLoad into EDR blocklists and threat intel platforms (MISP, Sentinel TAXII). AdaptixC2 payloads are frequently recompiled — treat hashes as point-in-time and prioritize behavior rules.
CVEsCVE-2025-64446, CVE-2026-0257, CVE-2019-6693, CVE-2021-42278/87Feed into vulnerability management SLA. CVE-2025-64446 and CVE-2026-0257 are actively exploited — treat as emergency patch items for any exposed FortiWeb/GlobalProtect instance.
Kit artifacts (non-IOC)Shared tagline, routing number, support phoneHigh-value for clustering: search brand-protection and Certificate Transparency logs (crt.sh) for newly registered domains matching kit naming patterns.

SOC teams should note that the phishing domains are disposable — blocking alone is insufficient. Ingest the domains into retro-hunts across 90 days of DNS/proxy logs to identify which users already submitted credentials, since those accounts require forced reset regardless of current block status.

Detection Engineering

YAML
---
title: k0t AdaptixC2 or MeshCentral RMM Persistence via Service Installation
id: 8f3a1c2e-4b7d-4e9a-b1f5-k0tadaptix0001
status: experimental
description: Detects installation of MeshCentral RMM agents or AdaptixC2 service persistence associated with the k0t edge-exploitation campaign. k0t deploys MeshCentral as a redundant C2 channel after FortiGate/GlobalProtect compromise.
author: Security Arsenal Threat Intelligence
date: 2026/11/22
references:
    - https://www.group-ib.com/blog/stray-cat-network-edge/
logsource:
    category: process_creation
    product: windows
detection:
    selection_meshcentral:
        Image|endswith:
            - '\meshagent.exe'
            - '\meshagent64.exe'
        CommandLine|contains:
            - 'girydsfsggvb'
            - '-install'
            - '-fullinstall'
    selection_sc:
        Image|endswith: '\sc.exe'
        CommandLine|contains|all:
            - 'create'
            - 'binpath'
        CommandLine|contains:
            - 'mesh'
            - 'adaptix'
            - 'update_svc'
    condition: 1 of selection_*
falsepositives:
    - Legitimate enterprise MeshCentral deployments (verify server FQDN against approved list)
level: high
tags:
    - attack.persistence
    - attack.t1543.003
    - attack.command_and_control
    - attack.t1219
---
title: Network Connection to k0t C2 or Fake-Bank Phishing Infrastructure
id: 7c2b9d4a-1e5f-4a8c-c3d2-k0tphish0002
status: experimental
description: Detects outbound connections to known k0t MeshCentral C2, 29-bank phishing kit domains, or GCP LLMjacking typosquat infrastructure from OTX pulses.
author: Security Arsenal Threat Intelligence
date: 2026/11/22
logsource:
    category: network_connection
    product: windows
detection:
    selection_domain:
        DestinationHostname|contains:
            - 'girydsfsggvb.com'
            - 'brayfinance.online'
            - 'firstcp.live'
            - 'securecapitalcorporationlimited.com'
            - 'qtbnk.com'
            - 'norexgo.com'
            - '1stggbinc.cc'
            - 'oakmontfb.com'
            - 'allyunitybn.com'
            - 'groosl.com'
            - 'groonl.com'
            - 'gboogl.com'
            - 'gbrokl.space'
            - 'gbrokl.xyz'
            - 'gcoalg.com'
            - 'gcslom.com'
            - 'gpoosl.com'
    filter_legit:
        Initiated: 'false'
    condition: selection_domain and not filter_legit
falsepositives:
    - Threat researchers or sandbox detonation environments
level: critical
tags:
    - attack.command_and_control
    - attack.t1071.001
    - attack.credential_access
    - attack.t1566
---
title: GCP Service Account Creation and External Invite — LLMjacking Pattern
id: 4d8e6f1b-9a3c-4b5d-d7e8-llmjack0003
status: experimental
description: Detects rapid sequence of GCP service account creation followed by org policy change or external member invitation — the LLMjacking attack chain used to hijack cloud billing for Gemini API abuse.
author: Security Arsenal Threat Intelligence
date: 2026/11/22
logsource:
    product: gcp
    service: cloudaudit
detection:
    selection_sa_create:
        methodName:
            - 'google.iam.admin.v1.CreateServiceAccount'
            - 'google.iam.admin.v1.CreateServiceAccountKey'
    selection_policy:
        methodName:
            - 'orgpolicy.policy.set'
            - 'SetIamPolicy'
        request.constraint|contains:
            - 'iam.allowedPolicyMemberDomains'
            - 'serviceaccounts.disableServiceAccountKeyCreation'
    selection_external:
        methodName: 'SetIamPolicy'
        request.bindings.members|contains:
            - 'gmail.com'
            - 'outlook.com'
    condition: 1 of selection_*
falsepositives:
    - Legitimate CI/CD service account provisioning; baseline on principal email and caller IP ASN
level: high
tags:
    - attack.persistence
    - attack.t1098.001
    - attack.defense_evasion
    - attack.t1562.007
KQL — Microsoft Sentinel / Defender
// Security Arsenal - k0t / AdaptixC2 / Phishing Kit / LLMjacking hunt
// Searches network, process, and identity telemetry for all three campaign clusters
let Lookback = 14d;
let C2Domains = dynamic([
    "girydsfsggvb.com", "brayfinance.online", "firstcp.live",
    "securecapitalcorporationlimited.com", "qtbnk.com", "norexgo.com",
    "1stggbinc.cc", "oakmontfb.com", "allyunitybn.com",
    "groosl.com", "groonl.com", "gboogl.com", "gbrokl.space",
    "gbrokl.xyz", "gcoalg.com", "gcslom.com", "gpoosl.com"
]);
let HashIOC = dynamic([
    "39c4decc4881ec02a47cd01c0fd8c3030527b466124ad03c6f99cfaa0ba0b72e",
    "6f5d18157425aa9697913ae1310f588f"
]);
let NetHits = DeviceNetworkEvents
    | where TimeGenerated > ago(Lookback)
    | where RemoteUrl has_any (C2Domains)
    | project TimeGenerated, DeviceName, InitiatingProcessAccountName,
              InitiatingProcessFileName, RemoteUrl, RemoteIP, ActionType;
let FileHits = DeviceFileEvents
    | where TimeGenerated > ago(Lookback)
    | where SHA256 in~ (HashIOC) or MD5 in~ (HashIOC)
    | project TimeGenerated, DeviceName, FileName, FolderPath, SHA256;
let RMMProc = DeviceProcessEvents
    | where TimeGenerated > ago(Lookback)
    | where FileName has_any ("meshagent", "meshagent64")
       or ProcessCommandLine has_any ("girydsfsggvb", "adaptix")
    | project TimeGenerated, DeviceName, AccountName, FileName,
              ProcessCommandLine, InitiatingProcessFileName;
let SuspiciousGPWLogons = SigninLogs
    | where TimeGenerated > ago(Lookback)
    | where AppDisplayName has_any ("GlobalProtect", "FortiGate", "FortiWeb")
       and ResultType != 0
    | summarize FailedAttempts = count(), DistinctIPs = dcount(IPAddress)
        by UserPrincipalName, bin(TimeGenerated, 1h)
    | where FailedAttempts > 20 or DistinctIPs > 5;
union NetHits, FileHits, RMMProc
| sort by TimeGenerated desc;
SuspiciousGPWLogons
| sort by FailedAttempts desc;
PowerShell
# Security Arsenal - k0t / AdaptixC2 / Phishing Kit IOC Hunt
# Run elevated. Scans for MeshCentral persistence, AdaptixC2 hashes,
# malicious service artifacts, and network beacons to campaign C2.

$Report = @()

# --- 1. MeshCentral / suspicious RMM services ---
$Services = Get-CimInstance Win32_Service | Where-Object {
    $_.PathName -match 'meshagent|adaptix' -or
    $_.Name -match 'mesh'
}
foreach ($svc in $Services) {
    $Report += [PSCustomObject]@{
        Category = 'Persistence-Service'
        Finding  = "$($svc.Name) | $($svc.PathName) | State=$($svc.State)"
        Severity = if ($svc.PathName -match 'girydsfsggvb|adaptix') {'Critical'} else {'Review'}
    }
}

# --- 2. Hash sweep of common payload staging paths ---
$Targets = @('39c4decc4881ec02a47cd01c0fd8c3030527b466124ad03c6f99cfaa0ba0b72e')
$MD5Targets = @('6f5d18157425aa9697913ae1310f588f')
$Paths = @("$env:TEMP", "$env:ProgramData", "C:\Users\Public", "C:\Windows\Temp")
foreach ($p in $Paths) {
    Get-ChildItem -Path $p -Recurse -File -ErrorAction SilentlyContinue |
        Where-Object { $_.Length -lt 50MB } | ForEach-Object {
            try {
                $h = Get-FileHash -Algorithm SHA256 -Path $_.FullName -ErrorAction Stop
                if ($Targets -contains $h.Hash.ToLower()) {
                    $Report += [PSCustomObject]@{Category='IOC-SHA256'; Finding=$_.FullName; Severity='Critical'}
                }
            } catch {}
        }
}

# --- 3. Active connections to campaign C2 / phishing infra ---
$BadDomains = @('girydsfsggvb.com','brayfinance.online','firstcp.live','qtbnk.com',
                'norexgo.com','groosl.com','gboogl.com','gcoalg.com')
Get-NetTCPConnection -State Established -ErrorAction SilentlyContinue | ForEach-Object {
    try {
        $rdns = ([System.Net.Dns]::GetHostEntry($_.RemoteAddress)).HostName
        foreach ($d in $BadDomains) {
            if ($rdns -match [regex]::Escape($d)) {
                $Report += [PSCustomObject]@{
                    Category='C2-Connection'
                    Finding="PID $($_.OwningProcess) -> $($_.RemoteAddress):$($_.RemotePort) ($rdns)"
                    Severity='Critical'
                }
            }
        }
    } catch {}
}

# --- 4. Rogue local admin accounts (k0t persistence via valid accounts) ---
$Admins = Get-LocalGroupMember -Group 'Administrators' -ErrorAction SilentlyContinue
foreach ($m in $Admins) {
    $Report += [PSCustomObject]@{
        Category='LocalAdmin-Audit'
        Finding=$m.Name
        Severity='Review'
    }
}

# --- 5. Run-key persistence referencing staging dirs ---
$RunKeys = @('HKLM:\SOFTWARE\Microsoft\Windows\CurrentVersion\Run',
             'HKCU:\SOFTWARE\Microsoft\Windows\CurrentVersion\Run')
foreach ($rk in $RunKeys) {
    Get-ItemProperty -Path $rk -ErrorAction SilentlyContinue |
        Select-Object * | Out-String | Select-String -Pattern 'meshagent|ProgramData|Public\\' -AllMatches |
        ForEach-Object {
            $Report += [PSCustomObject]@{Category='Persistence-RunKey'; Finding=$_.Line.Trim(); Severity='High'}
        }
}

$Report | Sort-Object Severity, Category | Format-Table -AutoSize
$Report | Export-Csv -Path ".\otx_hunt_$(Get-Date -Format 'yyyyMMdd_HHmm').csv" -NoTypeInformation
Write-Host "[+] Hunt complete. $($Report.Count) findings exported." -ForegroundColor Cyan

Response Priorities

Immediate (0–4 hours)

  • Block all listed domains at DNS, web proxy, and email gateway — including the full GCP typosquat set (groosl.com, gboogl.com, gcoalg.com, etc.) and the k0t MeshCentral C2 meshcentral.girydsfsggvb.com.
  • Load hashes 39c4decc...b72e (SHA256) and 6f5d1815...588f (MD5) into EDR blocklists.
  • Emergency patch or isolate any internet-facing FortiWeb (CVE-2025-64446) and Palo Alto GlobalProtect (CVE-2026-0257) instances; if patching is not possible today, place appliances behind VPN and disable management interfaces from WAN.
  • Run the KQL hunt across 14 days of telemetry; any MeshCentral agent or C2 connection is a confirmed intrusion — invoke IR immediately.
  • Audit GCP for service accounts and org-policy changes in the last 30 days; revoke unattested service account keys.

24 Hours

  • Force credential resets for any user whose browser resolved a fake-bank domain or GCP typosquat in proxy/DNS logs — assume harvested regardless of MFA status (AiTM kits defeat session tokens).
  • Rotate VPN/edge-device credentials globally if any FortiGate/GlobalProtect appliance was unpatched and internet-exposed during the campaign window. k0t's 70,000-credential haul means stale credentials are actively being sold or tested.
  • Review privileged GCP sign-ins for sponsored-ad referrers and anomalous ASN/geography; invalidate all active sessions for affected identities.
  • Retro-hunt authentication logs for brute-force patterns against edge appliances (>20 failures/hour or >5 source IPs per account).

1 Week

  • Architectural hardening of the edge: management interfaces off the public internet, certificate-based or phishing-resistant MFA (FIDO2) for all VPN and admin access, and network segmentation so a compromised edge appliance cannot reach internal AD (the CVE-2021-42278/42287 chain in k0t's toolkit targets exactly that pivot).
  • Cloud guardrails: enforce org-policy constraints that cannot be disabled by a single privileged user (require dual approval for iam.allowedPolicyMemberDomains changes), alert on any new service account key creation, and enable budget/anomaly alerts on Gemini API consumption to catch LLMjacking early.
  • Brand monitoring: register a CT-log monitoring pipeline for lookalike domains matching your banking/brand patterns; the 29-domain kit will redeploy.
  • EDR coverage gap review: edge appliances without agent coverage need syslog ingestion and NetFlow analytics as compensating detection.

Related Resources

Security Arsenal Incident Response Managed SOC & MDR Services AlertMonitor Threat Detection From The Dark Side Intel Hub

Is your security operations ready?

Get a free SOC assessment or see how AlertMonitor cuts through alert noise with automated triage.