Back to Intelligence

ARTEX Autonomous AI Agent (Storm-0940) + M365 AiTM Token Replay Campaign: OTX Pulse Analysis — Identity-Centric Detection Pack

SA
Security Arsenal Team
November 22, 2026
12 min read

Threat Summary

Two concurrent OTX pulses paint a single, coherent picture: enterprise identity infrastructure is the primary battlefield for late-2026 intrusion activity.

Pulse 1 — ARTEX in the Wild (Storm-0940): Throughout October 2026, operators aligned with Storm-0940, a Chinese state-nexus actor previously tied to supply-chain compromises, deployed ARTEX — an autonomous AI-driven penetration-testing agent — against IT and software vendors. ARTEX's targeting logic focused on identity providers: Okta, Microsoft Entra ID, and AWS Cognito. Observed tradecraft included reconnaissance of SaaS applications by parsing JavaScript bundles to map API endpoints and auth flows, followed by OAuth manipulation to establish durable, MFA-resistant access. Targeting software vendors is classic Storm-0940 supply-chain staging: compromise the vendor, inherit the customer base.

Pulse 2 — M365 Token Spray and Replay Campaign: Between July and September 2026, an unattributed but sophisticated actor ran an adversary-in-the-middle (AiTM) campaign against Microsoft 365 tenants. Phishing relays proxied live sign-in sessions, captured session cookies/tokens, and then replayed those tokens from infrastructure hosted by 1337 Services GmbH (Germany). The result: 79 organizations compromised, 497 successful replayed sign-ins over ten weeks — all bypassing MFA because the stolen artifacts were post-authentication session tokens.

Collective assessment: Both campaigns bypass the perimeter entirely by attacking the identity layer. One uses an autonomous agent to industrialize IdP reconnaissance and OAuth abuse against vendors (upstream access); the other weaponizes session theft at scale against end-user tenants (downstream access). Defenders should treat identity telemetry — not endpoint alerts — as ground truth for both.

Threat Actor / Malware Profile

Storm-0940 / ARTEX

  • Actor attribution: Chinese state-aligned threat actor tracked by Microsoft as Storm-0940; historically linked to supply-chain operations and password-spray activity against think tanks, government, and technology sectors.
  • Distribution / access method: ARTEX operates as an autonomous AI agent, not a conventional payload. It performs continuous reconnaissance — crawling public-facing JavaScript bundles of target web applications to enumerate hidden API routes, OAuth client IDs, and IdP configurations (Okta orgs, Entra tenants, Cognito user pools).
  • Payload behavior: Automates the full intrusion cycle against identity: discovers OAuth authorization endpoints, tests redirect URI misconfigurations, attempts consent phishing and token scope escalation, and manipulates OAuth grants to mint persistent access without triggering credential-based detections.
  • C2 communication: Rather than beaconing malware, ARTEX abuses legitimate IdP and SaaS APIs as its command channel — malicious OAuth tokens and API calls blend with normal application traffic over TLS to vendor domains.
  • Persistence: Rogue OAuth app consents and service principals in the victim tenant. These survive password resets and MFA enforcement — the defining trait of consent-based persistence.
  • Anti-analysis: The agent's autonomous behavior produces irregular, low-and-slow request patterns that defeat rate-based detections; traffic originates from rotating cloud infrastructure, and activity against IdPs is indistinguishable from legitimate app integration testing without behavioral baselines.

AiTM Token Replay Campaign (Unattributed)

  • Distribution: Phishing lures directing victims to AiTM relay pages that proxy the genuine Microsoft 365 login flow.
  • Payload behavior: The relay captures the post-authentication session cookie/token in real time — defeating MFA, including number-matching push and TOTP.
  • C2 / replay infrastructure: Stolen tokens replayed from IP space controlled by 1337 Services GmbH, a German hosting provider — a classic bulletproof-adjacent hosting pattern.
  • Persistence: Replayed sessions grant mailbox and SharePoint access; actors typically layer on inbox rules and may register additional auth methods or OAuth grants once inside, enabling business email compromise (BEC) follow-on operations.
  • Anti-analysis: Sessions replayed with the victim's legitimate token produce valid, non-anomalous authentication events; only network-origin analysis (impossible travel, datacenter ASN, session/IP mismatch) exposes the replay.

IOC Analysis

The pulses ship a limited indicator set — typical for TLP:WHITE community pulses where full IOCs live behind the referenced vendor reports. Available indicators:

TypeIndicatorContext
FileHash-SHA256d8a3fedb1a3cd1287ca47c6851a49e6052ab6116b43666e674765ca892d798ceAssociated with the M365 AiTM token replay campaign

Beyond atomic indicators, the highest-fidelity detection value in these pulses is behavioral and infrastructural:

  • Network-origin indicators: Sign-ins originating from hosting/VPN ASNs — specifically infrastructure attributed to 1337 Services GmbH — replaying Microsoft 365 sessions. SOC teams should enrich Entra sign-in logs with ASN classification rather than chasing IPs one by one.
  • Session-correlation indicators: The same Entra SessionId or token artifact appearing from two geographically or infrastructurally inconsistent IP addresses is a near-deterministic token-replay signature.
  • OAuth/consent indicators: New application consent grants, service principals with mail-read or offline_access scopes, and consent events to unverified publishers map directly to ARTEX-style persistence.

Operationalization guidance:

  1. Load the SHA256 into your EDR blocklist and retro-hunt (Sentinel DeviceFileEvents, CrowdStrike, Defender for Endpoint).
  2. Pull the referenced vendor reports (Wiz, Huntress) for expanded IP/domain sets and push them into your TI platform (MISP, Sentinel Threat Intelligence, or your SIEM lookup tables).
  3. Treat ASN- and behavior-based detections as primary; atomic IOCs expire fast against actors using rented hosting and autonomous tooling.
  4. Decode/pivot tooling: VirusTotal and Hybrid-Analysis for the hash; GreyNoise/AbuseIPDB for replay IPs; urlscan.io for AiTM relay infrastructure discovered in phishing takedowns.

Detection Engineering

YAML
---
title: Suspicious OAuth Consent Grant to Unverified Application
id: 7f3a1c2e-9b4d-4e1a-a8f6-2c5d9e1b7a30
status: experimental
description: Detects consent grants to unverified or newly registered applications with high-privilege scopes, consistent with ARTEX/Storm-0940 OAuth manipulation for tenant persistence.
references:
    - https://www.wiz.io/blog/artex-in-the-wild-storm-0940-expands-targeting-of-it-and-software-vendors
author: Security Arsenal Threat Intelligence
date: 2026/11/22
tags:
    - attack.persistence
    - attack.t1550.001
    - attack.t1528
logsource:
    product: azure
    service: auditlogs
detection:
    selection_operation:
        OperationName:
            - 'Consent to application'
            - 'Add service principal'
            - 'Add app role assignment to service principal'
    selection_scope:
        TargetResources|contains:
            - 'Mail.Read'
            - 'Mail.ReadWrite'
            - 'offline_access'
            - 'Directory.Read.All'
            - 'full_access_as_app'
    filter_verified:
        TargetResources|contains: 'verified'
    condition: selection_operation and selection_scope and not filter_verified
falsepositives:
    - Legitimate enterprise app onboarding by IT administrators
level: high
---
title: Microsoft 365 Sign-in from Datacenter Hosting ASN - Possible Token Replay
id: 2b8e4f1a-6c3d-4a9e-b5f7-8d1c3e6a9f24
status: experimental
description: Detects successful Entra ID interactive and non-interactive sign-ins originating from hosting provider address space, matching the AiTM token replay pattern observed from 1337 Services GmbH infrastructure.
references:
    - https://www.huntress.com/blog/token-replay-spray
author: Security Arsenal Threat Intelligence
date: 2026/11/22
tags:
    - attack.initial_access
    - attack.t1557
    - attack.t1539
    - attack.t1078.004
logsource:
    product: azure
    service: signinlogs
detection:
    selection:
        ResultType: '0'
    selection_network:
        IPAddress|startswith:
            - '193.'
            - '194.'
    keyword_hosting:
        NetworkLocationDetails|contains:
            - '1337 Services'
            - 'hosting'
            - 'datacenter'
    condition: selection and (selection_network or keyword_hosting)
falsepositives:
    - Corporate VPN egress through hosting providers
    - Remote users on cloud-hosted virtual desktops
level: medium
---
title: Session Token Reused Across Distant Network Origins
id: 4c9d2e7b-1a5f-4b8c-c6e3-7f2a4d8b1c59
status: experimental
description: Identifies the same Entra session identifier authenticating from multiple distinct IP addresses within a short window - a high-fidelity indicator of AiTM session cookie theft and replay.
references:
    - https://www.huntress.com/blog/token-replay-spray
author: Security Arsenal Threat Intelligence
date: 2026/11/22
tags:
    - attack.credential_access
    - attack.t1557
    - attack.t1539
logsource:
    product: azure
    service: signinlogs
detection:
    selection:
        SessionId|re: '.+'
    aggregation:
        - field: SessionId
          distinct: IPAddress
          gte: 2
    condition: selection and aggregation
falsepositives:
    - Mobile users roaming between cellular and Wi-Fi networks
level: critical
KQL — Microsoft Sentinel / Defender
// Hunt: AiTM Token Replay + Rogue OAuth Consent in Microsoft 365 Tenants
// Security Arsenal - OTX Pulse Analysis 2026-11

// Part 1: Same session token replayed from multiple IPs (token theft signature)
let ReplayWindow = 1h;
SigninLogs
| where TimeGenerated > ago(14d)
| where ResultType == 0
| where isnotempty(SessionId)
| summarize
    IPCount = dcount(IPAddress),
    IPs = make_set(IPAddress),
    Locations = make_set(Location),
    FirstSeen = min(TimeGenerated),
    LastSeen = max(TimeGenerated)
    by SessionId, UserPrincipalName, AppDisplayName
| where IPCount > 1 and LastSeen - FirstSeen <= ReplayWindow
| project UserPrincipalName, SessionId, AppDisplayName, IPs, Locations, FirstSeen, LastSeen
| order by LastSeen desc;

// Part 2: Non-interactive sign-ins from hosting/datacenter infrastructure
SigninLogs
| where TimeGenerated > ago(14d)
| where ResultType == 0
| where IPAddress !startswith "10." and IPAddress !startswith "192.168." and IPAddress !startswith "172."
| extend NetworkDetail = tostring(NetworkLocationDetails)
| where NetworkDetail has_any ("1337 Services", "hosting", "datacenter", "GmbH")
   or IPAddress in (dynamic([]))  // Populate with replay IPs from referenced vendor report
| project TimeGenerated, UserPrincipalName, IPAddress, NetworkDetail, AppDisplayName, ClientAppUsed, DeviceDetail
| order by TimeGenerated desc;

// Part 3: New OAuth consents with mail/directory scopes (ARTEX persistence pattern)
AuditLogs
| where TimeGenerated > ago(30d)
| where OperationName in ("Consent to application", "Add service principal", "Add app role assignment to service principal")
| mv-expand TargetResources
| extend ModifiedProps = tostring(TargetResources.modifiedProperties)
| where ModifiedProps has_any ("Mail.Read", "Mail.ReadWrite", "offline_access", "Directory.Read.All", "full_access_as_app")
| project TimeGenerated, OperationName, InitiatedBy = tostring(InitiatedBy.user.userPrincipalName),
          AppName = tostring(TargetResources.displayName), ModifiedProps, Result
| order by TimeGenerated desc;

// Part 4: Endpoint retro-hunt for campaign hash
DeviceFileEvents
| where TimeGenerated > ago(30d)
| where SHA256 == "d8a3fedb1a3cd1287ca47c6851a49e6052ab6116b43666e674765ca892d798ce"
| project TimeGenerated, DeviceName, FolderPath, FileName, InitiatingProcessAccountName;
PowerShell
# Security Arsenal - Identity Attack Hunt Script
# Targets: AiTM token replay artifacts + ARTEX-style OAuth persistence
# Requires: Microsoft.Graph PowerShell SDK (Connect-MgGraph -Scopes "AuditLog.Read.All","Directory.Read.All")
# Run on an admin workstation; endpoint checks can be remoted via Invoke-Command

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

Write-Host "[*] === Phase 1: Rogue OAuth App Consent Hunt (ARTEX persistence) ===" -ForegroundColor Cyan
Connect-MgGraph -Scopes "AuditLog.Read.All","Directory.Read.All" -NoWelcome

$consents = Get-MgAuditLogDirectoryAudit -Filter "activityDisplayName eq 'Consent to application'" -Top 200
foreach ($c in $consents) {
    $props = $c.TargetResources[0].ModifiedProperties
    if (($props | Out-String) -match 'Mail\.Read|offline_access|Directory\.Read\.All|full_access_as_app') {
        $report += [PSCustomObject]@{
            Phase      = 'OAuthConsent'
            Time       = $c.ActivityDateTime
            App        = $c.TargetResources[0].DisplayName
            Initiator  = $c.InitiatedBy.User.UserPrincipalName
            Risk       = 'HIGH - review app legitimacy and revoke if unauthorized'
        }
    }
}

Write-Host "[*] === Phase 2: Sign-in Replay Hunt (datacenter origins) ===" -ForegroundColor Cyan
$signins = Get-MgAuditLogSignIn -Top 500 -Filter "status/errorCode eq 0"
$replays = $signins | Group-Object -Property { $_.SessionId } | Where-Object {
    ($_.Group | Select-Object -ExpandProperty IPAddress -Unique).Count -gt 1
}
foreach ($r in $replays) {
    $report += [PSCustomObject]@{
        Phase     = 'TokenReplay'
        Time      = ($r.Group | Sort-Object CreatedDateTime)[-1].CreatedDateTime
        App       = ($r.Group)[0].AppDisplayName
        Initiator = ($r.Group)[0].UserPrincipalName
        Risk      = "CRITICAL - SessionId reused from: $(($r.Group | Select-Object -ExpandProperty IPAddress -Unique) -join ', ')"
    }
}

Write-Host "[*] === Phase 3: Endpoint Artifact Sweep (local host) ===" -ForegroundColor Cyan
$targetHash = 'd8a3fedb1a3cd1287ca47c6851a49e6052ab6116b43666e674765ca892d798ce'

# Hash hunt in common staging locations
$paths = @("$env:TEMP", "$env:APPDATA", "$env:LOCALAPPDATA", "C:\Users\Public", "C:\ProgramData")
foreach ($p in $paths) {
    Get-ChildItem -Path $p -Recurse -File -ErrorAction SilentlyContinue | ForEach-Object {
        if ((Get-FileHash $_.FullName -Algorithm SHA256).Hash -eq $targetHash) {
            $report += [PSCustomObject]@{ Phase='HashMatch'; Time=$_.LastWriteTime; App=$_.FullName; Initiator=$env:USERNAME; Risk='CRITICAL - campaign file found' }
        }
    }
}

# Suspicious persistence: Run keys and recently created scheduled tasks
$runKeys = @('HKLM:\SOFTWARE\Microsoft\Windows\CurrentVersion\Run',
             'HKCU:\SOFTWARE\Microsoft\Windows\CurrentVersion\Run')
foreach ($rk in $runKeys) {
    Get-ItemProperty $rk | ForEach-Object {
        $_.PSObject.Properties | Where-Object { $_.Value -match 'powershell|mshta|rundll32|wscript' } | ForEach-Object {
            $report += [PSCustomObject]@{ Phase='Persistence'; Time=(Get-Date); App="$rk -> $($_.Name)"; Initiator=$env:USERNAME; Risk="MEDIUM - review: $($_.Value)" }
        }
    }
}
Get-ScheduledTask | Where-Object { $_.Date -gt (Get-Date).AddDays(-14) -and $_.TaskPath -notlike '\Microsoft*' } | ForEach-Object {
    $report += [PSCustomObject]@{ Phase='Persistence'; Time=$_.Date; App=$_.TaskName; Initiator=$_.Principal.UserId; Risk='MEDIUM - recently created task, validate' }
}

# Active connections to non-corporate hosting ranges
Get-NetTCPConnection -State Established | Where-Object { $_.RemotePort -in 443,8443 -and $_.RemoteAddress -notmatch '^(10\.|192\.168\.|172\.)' } |
    ForEach-Object {
        $proc = (Get-Process -Id $_.OwningProcess).ProcessName
        $report += [PSCustomObject]@{ Phase='Network'; Time=(Get-Date); App="$proc -> $($_.RemoteAddress):$($_.RemotePort)"; Initiator=$env:USERNAME; Risk='INFO - enrich remote IP against replay infra TI' }
    }

$report | Export-Csv -Path ".\IdentityHunt_$(Get-Date -Format 'yyyyMMdd_HHmm').csv" -NoTypeInformation
$report | Format-Table -AutoSize
Write-Host "[+] Hunt complete. Results exported." -ForegroundColor Green

Response Priorities

Immediate (0-4 hours)

  • Block and retro-hunt the campaign hash d8a3fedb1a3cd1287ca47c6851a49e6052ab6116b43666e674765ca892d798ce across EDR and email gateways; pull expanded IP/domain sets from the referenced Wiz and Huntress reports into blocklists.
  • Run the session-replay KQL query across all Entra tenants. Any SessionId seen from multiple IPs = confirmed token theft: revoke the session, force credential reset, and re-register MFA for the affected user.
  • Audit OAuth consents tenant-wide. Revoke any unverified third-party app grants carrying Mail.Read, offline_access, or directory scopes created in the last 90 days.
  • If your organization is an IT/software vendor, assume upstream targeting: enumerate service principals, CI/CD tokens, and marketplace app integrations for signs of ARTEX-style probing.

24 Hours

  • Identity verification wave: Because both campaigns defeat MFA at the session layer, do not trust MFA status as proof of integrity. Re-authenticate affected users from known-good devices, revoke all refresh tokens (Revoke-MgUserSignInSession), and review inbox rules, forwarding, and delegate access for BEC staging.
  • Review Entra app registrations and service principals created or modified since September 2026; validate business justification for each.
  • For software vendors: rotate signing keys, audit build-system access logs, and notify downstream customers if any artifact tampering is found — Storm-0940's objective is your customer list.
  • Deploy the Sigma rules above into production and tune Part 2 of the KQL against your legitimate VPN/VDI egress ranges.

1 Week

  • Adopt token-binding controls: migrate priority users to phishing-resistant auth (FIDO2/passkeys) and enable Continuous Access Evaluation (CAE) plus Conditional Access token protection (where licensing allows) so stolen cookies replay as invalid.
  • Harden OAuth surface: disable user consent for third-party apps, require admin consent workflows, and enable Entra app consent policies restricting scopes.
  • Segment CI/CD and package-registry access for software vendors; enforce workload identity federation with short-lived tokens instead of standing secrets.
  • Add ASN-based detections to your SIEM: alert on any M365/IdP sign-in from classified hosting-provider address space.
  • Tabletop the combined scenario: an upstream vendor compromise (ARTEX pattern) chained into downstream AiTM session theft against your tenant — validate IR runbooks cover both halves.

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.