Back to Intelligence

Aoki Infostealer Targets AI Coding Assistant Credentials + CoGUI Phishing Kit Disruption: OTX Pulse Analysis — Enterprise Detection Pack

SA
Security Arsenal Team
October 15, 2026
10 min read

This briefing synthesizes four concurrent AlienVault OTX pulses (TLP:WHITE, 2026-10-15) that collectively paint a clear picture: credential theft is industrializing, and adversaries are pivoting toward high-value identity artifacts — AI assistant API keys, session tokens, and 2FA-bypassed social logins — while simultaneously adopting AI agents to automate intrusion execution.


Threat Summary

Four distinct but thematically linked activity clusters emerged this cycle:

  1. Aoki (Aokigahara) Infostealer has added a novel theft module targeting AI coding assistants — Claude, OpenAI Codex, Cursor, Codeium, and Gemini. This is a first-of-its-kind expansion: earlier builds only targeted browsers, crypto wallets, and standard session stores. Stolen API keys and session files let attackers operate under legitimate developer identities and access connected repositories, CI/CD pipelines, and cloud resources.

  2. Fake Calendly "Meta Careers" phishing kit targets job seekers with spoofed interview-booking pages. Victims complete fake CAPTCHAs and are funneled into Facebook and Google credential capture with two-factor authentication bypass — likely a real-time reverse-proxy (AiTM) component harvesting session cookies.

  3. First confirmed LLM-agent-driven intrusion: an exposed, unauthenticated Spring Batch job-submission endpoint on an Apache Tomcat server was exploited end-to-end by Cairn, an open-source AI agent orchestration platform. No zero-days, no malware — the agents executed reconnaissance, exploitation, and privilege escalation autonomously within 24 hours.

  4. YAM Store disruption: a coordinated takedown hit Chinese-speaking operators behind the CoGUI, CoAceV, and CoFish_Ws phishing kits, which pushed hundreds of millions of phishing messages primarily at Japanese organizations, impersonating 66 consumer and finance brands across finance, healthcare, government, education, and manufacturing verticals in Japan, the US, Australia, and Italy.

Collective signal: the common denominator is credential and identity abuse. Whether via endpoint infostealers, AiTM phishing, or autonomous AI exploitation of exposed endpoints, adversaries are monetizing access tokens faster than traditional password-centric defenses can respond. The YAM Store disruption will cause short-term kit fragmentation — expect rebranded CoGUI successors and domain churn within weeks.


Threat Actor / Malware Profile

Aoki / Aokigahara Infostealer

  • Attribution: Unknown operator; malware-as-a-service distribution model suspected based on hash diversity (27 indicators, multiple SHA256 samples).
  • Distribution: Typically delivered via cracked software, malvertising, and trojanized developer tooling — consistent with the new AI-assistant targeting, which implies a developer/user base deliberately selected for high-value tokens.
  • Payload behavior: Enumerates browser credential stores, cryptocurrency wallet files, and — new in this build — configuration and session directories for Claude, Codex, Cursor, Codeium, and Gemini clients. Extracts API keys, OAuth tokens, and session files, stages them in a compressed archive, and exfiltrates over HTTPS.
  • C2: Observed infrastructure includes the domain neihite.cc. Communication is HTTPS POST-based with encrypted payloads, blending with legitimate TLS traffic.
  • Persistence: Commonly establishes Run-key registry persistence and may drop scheduled tasks for re-execution after reboot.
  • Anti-analysis: Packing/obfuscation across samples (evidenced by multiple distinct SHA256 hashes), sandbox-aware sleep routines, and string encryption.

Fake Calendly / Meta Careers Phishing Kit

  • Attribution: Unknown; kit-based operation using join*.cfd/.works/.info domain clusters.
  • Delivery: Job-scam lures via email and social platforms impersonating Meta recruiting.
  • Behavior: Multi-stage social engineering — time-slot selection, PII capture, fake CAPTCHA — before AiTM-style credential capture against Facebook and Google, including 2FA session interception.

Cairn AI-Agent Intrusion

  • Attribution: Unknown actor leveraging the open-source Cairn orchestration platform.
  • Attack chain: Unauthenticated Spring Batch job submission on Apache Tomcat → arbitrary job execution → LLM-agent-driven enumeration and privilege escalation → full server takeover within 24 hours.
  • Infrastructure: 204.194.54.240, 204.194.55.189, 94.177.131.113.

YAM Store / CoGUI-CoAceV-CoFish_Ws Kits

  • Attribution: Chinese-speaking cybercriminal collective tracked as YAM Store.
  • Scale: Hundreds of millions of phishing messages; 66 impersonated brands; heavy Japan focus (e.g., kumamoto-city.jp-spoofing URL paths, Paidy payment brand lures).
  • Status: Disrupted via domain seizure — expect infrastructure regeneration.

IOC Analysis

The pulse set contains four indicator classes, each requiring a different operationalization path:

TypeExamplesOperationalization
FileHash-SHA256 (Aoki)bf037920...caebf, 752c8e64...4b69Push to EDR blocklists and SIEM threat-intel lookups; hashes burn fast — prioritize behavioral detections below
Domains (Aoki C2 / phishing kits)neihite.cc, joinconnect.cfd, joinhire.info, lygyaou.com, kenandy.shopDNS sinkhole/block at resolver and proxy; alert on historical resolution in DNS logs
URLs (YAM Store kit pages)https://paidy.farradae.com/paidy/ubmissionsjp, ...kumamoto-city.jp pathsURL-regex blocking at secure web gateway; hunt proxy logs for the distinctive path patterns (ubmissionsjp, kumamoto-city.jp suffixes)
IPv4 (Cairn intrusion infra)204.194.54.240, 94.177.131.113Block at perimeter; retro-hunt netflow, firewall, and Tomcat access logs for inbound contact

Tooling: MISP or OpenCTI for IOC ingestion and correlation; Suricata/Zeek for network-side domain and IP matching; YARA for hash-adjacent sample clustering; sandbox detonation (ANY.RUN, Hybrid Analysis) for new Aoki variants given the rapid hash rotation.


Detection Engineering

YAML
---
title: Aoki Infostealer AI Assistant Credential Access
id: 9f1e2a7c-4b3d-4e5f-8a1b-aoki0001c001
status: experimental
description: Detects processes accessing AI coding assistant credential, session, or configuration stores targeted by the Aoki/Aokigahara infostealer (Claude, Codex, Cursor, Codeium, Gemini).
author: Security Arsenal Threat Intelligence
references:
  - https://www.security.com/threat-intelligence/aoki-infostealer-ai-targeting
date: 2026/10/15
modified: 2026/10/15
tags:
  - attack.credential_access
  - attack.t1555
  - attack.t1552.001
logsource:
  category: file_event
  product: windows
detection:
  selection_paths:
    TargetFilename|contains:
      - '\.cursor\'
      - '\Cursor\User\'
      - '\.claude\'
      - '\.config\claude\'
      - '\.codex\'
      - '\.codeium\'
      - '\.gemini\'
      - '\Google\Gemini\'
  selection_tokens:
    TargetFilename|contains:
      - 'session'
      - 'token'
      - 'api_key'
      - 'credentials'
      - 'auth'
  filter_legit:
    Image|contains:
      - '\Cursor\'
      - '\Codeium\'
      - '\Anthropic\'
      - '\OpenAI\'
  condition: selection_paths and selection_tokens and not filter_legit
falsepositives:
  - AI assistant applications updating their own session files
  - Legitimate developer backup scripts
level: high
---
title: AiTM Phishing Domain Cluster Resolution - Fake Calendly and YAM Store Kits
id: 8d2b3c4e-5a6f-4b7c-9d0e-phish0002c02
status: experimental
description: Detects DNS resolution of known fake Calendly Meta Careers phishing domains and YAM Store (CoGUI/CoAceV/CoFish_Ws) kit infrastructure, plus TLD patterns associated with the campaign.
author: Security Arsenal Threat Intelligence
references:
  - https://www.malwarebytes.com/blog/threat-intel/2026/10/fake-calendly-meta-careers-pages-steal-facebook-and-google-logins
  - https://www.proofpoint.com/us/blog/threat-insight/disruption-targets-major-chinese-language-phishing-kits-japanese-focus
date: 2026/10/15
modified: 2026/10/15
tags:
  - attack.initial_access
  - attack.t1566.002
  - attack.t1557
logsource:
  category: dns
detection:
  selection_domains:
    query|contains:
      - 'joinconnect.'
      - 'joinhire.'
      - 'joinquest.'
      - 'jointhetalent.'
      - 'lygyaou.com'
      - 'kenandy.shop'
      - 'new-app-haixing.com'
      - 'farradae.com'
      - 'ylhgkj.com'
      - 'neihite.cc'
  condition: selection_domains
falsepositives:
  - Threat researcher investigation
level: critical
---
title: Suspicious Spring Batch Job Submission on Tomcat - AI Agent Exploitation Pattern
id: 7c1a0b9d-2e3f-4a5b-8c7d-cairn003c003
status: experimental
description: Detects process spawning from Apache Tomcat consistent with exploitation of an exposed unauthenticated Spring Batch job-submission endpoint, as observed in the Cairn AI-agent-driven server takeover.
author: Security Arsenal Threat Intelligence
references:
  - https://reliaquest.com/blog/threat-spotlight-how-ai-agents-turned-one-exposed-endpoint-into-a-server-takeover
date: 2026/10/15
modified: 2026/10/15
tags:
  - attack.initial_access
  - attack.t1190
  - attack.privilege_escalation
logsource:
  category: process_creation
  product: windows
detection:
  selection_parent:
    ParentImage|endswith:
      - '\java.exe'
      - '\tomcat9.exe'
      - '\tomcat8.exe'
      - '\catalina.exe'
  selection_child:
    Image|endswith:
      - '\cmd.exe'
      - '\powershell.exe'
      - '\pwsh.exe'
      - '\whoami.exe'
      - '\net.exe'
      - '\curl.exe'
      - '\certutil.exe'
  condition: selection_parent and selection_child
falsepositives:
  - Legitimate Spring Batch administrative jobs
  - Application deployment automation
level: high
KQL — Microsoft Sentinel / Defender
// Security Arsenal Hunt: Aoki C2, phishing kit resolution, and Cairn intrusion infrastructure
let SuspiciousDomains = dynamic(["neihite.cc", "joinconnect.cfd", "joinconnect.works", "joinhire.cfd", "joinhire.info", "joinquest.cfd", "joinquest.info", "joinquest.works", "jointhetalent.cfd", "lygyaou.com", "kenandy.shop", "new-app-haixing.com", "farradae.com", "ylhgkj.com"]);
let CairnIPs = dynamic(["204.194.54.240", "204.194.55.189", "94.177.131.113"]);
union isfuzzy=true
(
    DeviceNetworkEvents
    | where TimeGenerated > ago(14d)
    | where RemoteUrl has_any (SuspiciousDomains) or RemoteIP in (CairnIPs)
    | project TimeGenerated, DeviceName, InitiatingProcessFileName, InitiatingProcessCommandLine, RemoteIP, RemoteUrl, ActionType
),
(
    DeviceProcessEvents
    | where TimeGenerated > ago(14d)
    | where SHA256 in ("bf037920d39b85572ec7e53b51f09c28198c2c0afbc7c7d6cf639b1f047caebf", "752c8e64af60960b85cf00e4bd34eb459e24b7939464f718ec18494df43b4b69", "0adc8e83b2416c275b1283916d70568faddeb6d718eb45f9addfb0938db245b3", "6655b89b9885483d459c73f2a52e3cec121de5c5956172a8676a0f4c2de94999", "780bbdbb119befd4eb16e0c2ec0c64e7bedc1b53ce4f85804aea091a6e3af519", "2fb80e2221bd304e7cd5ac04169e71807ad92d19836f1aaf7885295cb90b2d94", "73dbac0907e8a092cd978f2787ab1994946f5da1e08f2bdbd05ea52e80cdb896")
    | project TimeGenerated, DeviceName, FileName, FolderPath, SHA256, ProcessCommandLine, InitiatingProcessFileName
)
| sort by TimeGenerated desc
PowerShell
# Security Arsenal IOC Hunt — Aoki Infostealer & AI Credential Theft Artifacts
# Run elevated on suspected endpoints. Outputs findings to C:\IR\AokiHunt_$timestamp.csv

$ErrorActionPreference = 'SilentlyContinue'
$findings = @()

# 1. Registry Run-key persistence check
$runKeys = @(
    'HKCU:\Software\Microsoft\Windows\CurrentVersion\Run',
    'HKLM:\Software\Microsoft\Windows\CurrentVersion\Run',
    'HKCU:\Software\Microsoft\Windows\CurrentVersion\RunOnce'
)
foreach ($key in $runKeys) {
    Get-ItemProperty -Path $key | ForEach-Object {
        $_.PSObject.Properties | Where-Object { $_.Name -notmatch '^PS' } | ForEach-Object {
            $findings += [PSCustomObject]@{ Type='RunKey'; Path=$key; Name=$_.Name; Value=$_.Value }
        }
    }
}

# 2. AI assistant credential/session artifact access staging
$aiPaths = @(
    "$env:USERPROFILE\.claude", "$env:USERPROFILE\.codex",
    "$env:USERPROFILE\.codeium", "$env:USERPROFILE\.gemini",
    "$env:APPDATA\Cursor", "$env:APPDATA\Codeium"
)
foreach ($p in $aiPaths) {
    if (Test-Path $p) {
        Get-ChildItem -Path $p -Recurse -File | Where-Object {
            $_.Name -match 'session|token|key|cred|auth' -and
            $_.LastAccessTime -gt (Get-Date).AddDays(-3) -and
            $_.LastAccessTime -gt $_.LastWriteTime
        } | ForEach-Object {
            $findings += [PSCustomObject]@{ Type='SuspiciousAIFileAccess'; Path=$_.FullName; Name=$_.Name; Value="Accessed $($_.LastAccessTime)" }
        }
    }
}

# 3. Known Aoki SHA256 hash sweep of common staging dirs
$aokiHashes = @('bf037920d39b85572ec7e53b51f09c28198c2c0afbc7c7d6cf639b1f047caebf',
'752c8e64af60960b85cf00e4bd34eb459e24b7939464f718ec18494df43b4b69',
'0adc8e83b2416c275b1283916d70568faddeb6d718eb45f9addfb0938db245b3',
'6655b89b9885483d459c73f2a52e3cec121de5c5956172a8676a0f4c2de94999',
'780bbdbb119befd4eb16e0c2ec0c64e7bedc1b53ce4f85804aea091a6e3af519',
'2fb80e2221bd304e7cd5ac04169e71807ad92d19836f1aaf7885295cb90b2d94',
'73dbac0907e8a092cd978f2787ab1994946f5da1e08f2bdbd05ea52e80cdb896')
foreach ($dir in @("$env:TEMP", "$env:APPDATA", "$env:LOCALAPPDATA\Temp", "$env:USERPROFILE\Downloads")) {
    Get-ChildItem -Path $dir -Recurse -File -Depth 2 | ForEach-Object {
        $h = (Get-FileHash -Algorithm SHA256 $_.FullName).Hash.ToLower()
        if ($aokiHashes -contains $h) {
            $findings += [PSCustomObject]@{ Type='AokiHashMatch'; Path=$_.FullName; Name=$_.Name; Value=$h }
        }
    }
}

# 4. Active connections to known C2 / phishing infrastructure
Get-NetTCPConnection -State Established | Where-Object {
    $_.RemoteAddress -in @('204.194.54.240','204.194.55.189','94.177.131.113')
} | ForEach-Object {
    $proc = Get-Process -Id $_.OwningProcess
    $findings += [PSCustomObject]@{ Type='MaliciousConnection'; Path=$proc.Path; Name=$proc.Name; Value="$($_.RemoteAddress):$($_.RemotePort)" }
}

# 5. Scheduled tasks created in the last 7 days (persistence)
Get-ScheduledTask | Where-Object { $_.Date -and ([datetime]$_.Date) -gt (Get-Date).AddDays(-7) } | ForEach-Object {
    $findings += [PSCustomObject]@{ Type='RecentScheduledTask'; Path=$_.TaskPath; Name=$_.TaskName; Value=$_.State }
}

New-Item -Path 'C:\IR' -ItemType Directory -Force | Out-Null
$out = "C:\IR\AokiHunt_$(Get-Date -Format 'yyyyMMdd_HHmmss').csv"
$findings | Export-Csv -Path $out -NoTypeInformation
Write-Host "[+] Hunt complete. $($findings.Count) findings written to $out"

Response Priorities

Immediate (0–4 hours)

  • Block all pulse IOCs at DNS resolver, secure web gateway, and EDR: neihite.cc, the full join* phishing cluster, YAM Store kit domains/URLs, and the three Cairn-associated IPv4 addresses.
  • Push the seven Aoki SHA256 hashes to EDR prevention policy; deploy the Sigma rules above.
  • Retro-hunt 14 days of DNS, proxy, and netflow telemetry for the listed domains and IPs using the KQL query.
  • Audit externally exposed Apache Tomcat/Spring Batch endpoints; take unauthenticated job-submission interfaces offline or behind authentication now.

24 Hours (Credential Compromise Window)

  • Because Aoki steals AI-assistant API keys and session tokens, force rotation of all developer-held AI service credentials (Claude, OpenAI, Cursor, Codeium, Gemini) for any endpoint flagged in the hunt — plus GitHub/GitLab tokens reachable from those machines.
  • Revoke active sessions and OAuth grants for Facebook and Google Workspace accounts of any user who interacted with the fake Calendly domains; AiTM phishing captures session cookies, so password reset alone is insufficient.
  • Audit cloud and repository audit logs for API-key usage from anomalous source IPs.

1 Week (Architecture Hardening)

  • Move AI assistant credentials out of on-disk plaintext stores into a secrets manager (HashiCorp Vault, AWS Secrets Manager) with short-lived tokens.
  • Deploy FIDO2/phishing-resistant MFA — the Calendly kit's 2FA bypass targets TOTP and push-based factors; hardware keys defeat AiTM relay.
  • Enforce egress filtering to block low-reputation TLDs (.cfd, .works, .info) for standard users and flag .cc lookups for review.
  • Implement mTLS and authentication on all internal job-submission and orchestration endpoints; assume AI-agent-driven attackers can exploit any unauthenticated interface at machine speed.
  • Subscribe to YAM Store successor-kit tracking — the CoGUI/CoAceV/CoFish_Ws disruption will regenerate under new brands within weeks.

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.