Back to Intelligence

LLMJacking UNC6993 + AI Phishing Ops: OTX Pulse Analysis — Credential Theft Detection Pack

SA
Security Arsenal Team
November 3, 2026
16 min read

Security teams are facing a coordinated surge in credential theft operations that converge on artificial intelligence infrastructure, identity systems, and financial payment channels. Analysis of four AlienVault OTX pulses published between 2026-11-02 and 2026-11-03 reveals a threat landscape where adversaries are no longer merely harvesting passwords — they are monetizing access to enterprise Large Language Model deployments, intercepting one-time passcodes in real-time, and leveraging trusted collaboration platforms to escalate directly into Active Directory domain dominance.

This briefing synthesizes live OTX pulse data to deliver actionable detection engineering, threat hunting queries, and response priorities for enterprise security operations centers.


Threat Summary

The four OTX pulses analyzed collectively expose a multi-vector credential theft ecosystem that treats AI infrastructure as a high-value financial target. This is not opportunistic cryptomining — it is systematic financial fraud through three converging attack chains.

LLMJacking Campaign (UNC6993) represents the most financially damaging operation. Threat actors compromise cloud environments and hijack enterprise access to LLM APIs, generating fraudulent usage charges exceeding $100,000 per incident. Unlike cryptojacking's gradual resource consumption, LLMJacking creates immediate, massive billing spikes. The campaign targets financial services, retail, technology, and transportation sectors across North America, South America, and Asia Pacific. Associated malware includes Bankshot (S0239) and Trojan Manuscript, with exploitation of CVE-2026-40217 for initial access or privilege escalation. Infrastructure includes domains such as jet-night.com and newapi.pro, which likely serve as proxy endpoints or credential harvesting panels for stolen API keys.

AI Billing Phishing Operations run parallel infrastructure targeting end users. Fraudulent ChatGPT Plus billing emails direct victims to compromised WordPress instances hosted on Nexcess's nxcli.io platform. The operation features live attacker involvement — operators monitor submitted credit card details in real-time and dynamically request OTP codes to bypass multi-factor authentication. This indicates a mature carding operation with immediate monetization capability, not merely credential resale.

Microsoft Teams AD Hijacking demonstrates sophisticated social engineering for enterprise privilege escalation. Attackers impersonate IT help desk personnel, convincing users to launch Quick Assist remote support sessions. Once inside, they deploy portable Python, Bun, and Java runtimes to enumerate Active Directory environments. The campaigns leverage PetitPotam (CVE-2021-36942 or related NTLM relay techniques) for domain controller compromise, ESC1 and ESC8 certificate service misconfigurations for privilege escalation, and Rclone for data exfiltration. The use of DCSync attacks indicates successful domain administrator compromise.

AI Developer Platform Breaches affected at least nine platforms in October 2026: Fireworks AI, Andromeda, Zeabur, Lightning AI, OpenCode, Railway, Warp, CommandCode, and Northflank. Attackers exploited vulnerabilities to access stored environment variables containing AI credentials, API tokens, and cloud access keys. Infrastructure includes IPv4 addresses 31.57.201.201 and 38.246.114.118, plus hostnames artex.nhhh.de and direct.465415.xyz.

The collective objective is financial gain through three mechanisms: direct cloud billing fraud via LLMJacking, payment card theft via phishing, and ransomware/extortion preparation via AD compromise. The supply chain dimension of AI developer platform breaches suggests downstream targeting of organizations that integrated these platforms.


Threat Actor / Malware Profile

UNC6993 — LLMJacking Campaign Operator

Attribution: AlienVault tracks this cluster as UNC6993, indicating an uncategorized but distinctly clustered threat activity group. The consistent targeting of cloud LLM infrastructure and the specific tooling suggest a financially motivated actor with cloud-native expertise, likely operating from jurisdictions with limited law enforcement cooperation.

Malware Families:

Bankshot (S0239) is a remote access trojan historically associated with North Korean state actors (Lazarus Group), but its appearance in UNC6993 infrastructure suggests either tool sharing, false flag operations, or independent acquisition. Bankshot provides persistent access, credential harvesting, and lateral movement capabilities. Its presence in LLMJacking operations may indicate use for initial cloud console compromise rather than direct API theft.

Trojan Manuscript is less documented in public sources, suggesting a custom or rebranded payload. Based on LLMJacking operational requirements, it likely functions as a credential stealer targeting cloud provider configuration files (~/.aws/credentials, ~/.azure/, gcloud credentials), environment variables, and browser-stored session tokens for cloud consoles.

Distribution Method: Initial access likely occurs through exploitation of CVE-2026-40217 (details not fully disclosed in OTX), phishing, or compromised credentials from the AI developer platform breaches. The "manus" tag may indicate manual attacker operations rather than automated deployment.

Payload Behavior: LLMJacking malware prioritizes theft of AI API keys (OpenAI, Anthropic, Azure OpenAI, AWS Bedrock), cloud service account credentials, and SSH keys. Unlike traditional stealers, it includes logic to test key validity and quota limits, transmitting only high-value, active credentials to C2 infrastructure.

C2 Communication: Observed domains (jet-night.com, newapi.pro) use HTTPS with domain fronting or legitimate CDN services to blend with normal traffic. The "llm mixers" tag suggests infrastructure that proxies stolen API requests through multiple accounts to obscure attribution and evade rate limiting.

Persistence: Cloud-native persistence mechanisms include creation of additional service accounts, IAM policy modifications, Lambda/Cloud Functions implants, and modification of container images in registries. Traditional endpoint persistence uses scheduled tasks and WMI event subscriptions.

Anti-Analysis: Bankshot historically employs code obfuscation, anti-debugging checks, and VM detection. Trojan Manuscript likely checks for cloud security agent presence (CrowdStrike, Wiz, Orca) before executing credential theft routines.

AI Billing Phishing Infrastructure

Delivery: Compromised WordPress instances on Nexcess nxcli.io hosting platform. Specific observed hosts: 0b8918640c.nxcli.io, c54b986c1c.nxcli.io.

Social Engineering: Urgent billing failure notifications for ChatGPT Plus subscriptions create artificial time pressure. The use of legitimate-appearing domains (authgpt.info, ai-billing.info, url-y.uk) bypasses basic domain reputation filters.

Live Operator Involvement: The most distinctive feature is real-time OTP interception. When victims submit payment details, attackers immediately attempt card verification or fraudulent transactions, triggering legitimate OTP requests that are then requested from the victim. This requires 24/7 operator availability and indicates organized crime involvement.

Teams AD Hijacking TTPs

Initial Access: Microsoft Teams external messaging or guest account abuse to initiate contact with target users.

Execution: Quick Assist (Microsoft's remote assistance tool) provides legitimate remote access without malware deployment, evading many EDR solutions.

Privilege Escalation: PetitPotam exploits NTLM relay vulnerabilities against Active Directory Certificate Services (AD CS). ESC1 exploits certificate templates that allow client authentication and domain controller certificate requests. ESC8 exploits HTTP-based certificate enrollment interfaces.

Collection: Portable Python, Bun, and Java runtimes enable execution of enumeration scripts without installing software, leaving minimal forensic artifacts.

Exfiltration: Rclone with encrypted configuration files transfers data to cloud storage (Mega, Dropbox, or actor-controlled S3 buckets).


IOC Analysis

The indicator set spans network infrastructure, file hashes, and vulnerability identifiers, requiring differentiated operationalization strategies.

Network Indicators

TypeValueContextAction
IPv431.57.201.201AI platform breach C2Block egress; investigate any historical connections
IPv438.246.114.118AI platform breach C2Block egress; investigate any historical connections
Domainjet-night.comLLMJacking C2DNS sinkhole; TLS fingerprinting for JA3/JA3S matching
Domainnewapi.proLLMJacking C2DNS sinkhole; monitor for subdomain enumeration
Domainauthgpt.infoPhishingBlock; add to web proxy category
Domainai-billing.infoPhishingBlock; add to web proxy category
Domainurl-y.ukPhishingBlock; URL shortener monitoring
Hostname0b8918640c.nxcli.ioPhishingBlock; coordinate with Nexcess abuse team
Hostnamec54b986c1c.nxcli.ioPhishingBlock; coordinate with Nexcess abuse team
Hostnameartex.nhhh.deAI platform breachBlock; investigate for credential exfiltration
Hostnamedirect.465415.xyzAI platform breachBlock; DGA-like pattern suggests flux infrastructure

File Hash Indicators

Eight file hashes across SHA256, MD5, and SHA1 algorithms represent portable runtime executables and enumeration scripts. The presence of multiple hash algorithms for identical files suggests detection evasion testing by adversaries. SOC teams should implement all hashes in EDR blocklists and conduct retroactive hunts across email gateways, web proxies, and file shares.

Vulnerability Indicator

CVE-2026-40217 requires immediate vulnerability scanner correlation. Given the LLMJacking context, this likely affects cloud management interfaces, AI platform SDKs, or container orchestration systems. Prioritize internet-facing assets and cloud management plane access.

Operationalization Workflow

  1. Ingestion: Import OTX pulses into TIP (Threat Intelligence Platform) with confidence scoring
  2. Correlation: Match against internal DNS, proxy, EDR, and cloud flow logs for 90-day retroactive analysis
  3. Enrichment: Query passive DNS for subdomain infrastructure; WHOIS for registrant patterns
  4. Action: Network blocks for active C2; hash blocks for execution prevention; ticket for vulnerability remediation
  5. Feedback: Report false positives to OTX; contribute newly observed infrastructure

Detection Engineering

Sigma Rules

YAML
---
title: LLMJacking Credential Access via Cloud Configuration Files
id: 7a3b9c2d-8e4f-4a1b-9c3d-5e6f7a8b9c0d
status: experimental
description: Detects access to cloud provider credential files consistent with LLMJacking operations targeting AI API keys and cloud service accounts
author: Security Arsenal Threat Intelligence
date: 2026/11/03
modified: 2026/11/03
tags:
    - attack.credential_access
    - attack.t1552
    - attack.t1552.001
logsource:
    category: file_access
    product: windows
    definition: 'Requires Sysmon or similar file access auditing'
detection:
    selection_paths:
        TargetFilename|contains:
            - '\.aws\credentials'
            - '\.azure\'
            - '\.config\gcloud\'
            - '\.kube\config'
            - '\.docker\config.json'
    selection_process:
        Image|endswith:
            - '\python.exe'
            - '\python3.exe'
            - '\bun.exe'
            - '\java.exe'
            - '\node.exe'
            - '\powershell.exe'
            - '\cmd.exe'
    condition: selection_paths and selection_process
falsepositives:
    - DevOps automation scripts accessing cloud credentials
    - Cloud SDK installations
level: high
---
title: PetitPotam NTLM Relay and AD CS Exploitation
id: 8b4c0d3e-9f5a-5b2c-0d4e-6f7a8b9c0d1e
status: experimental
description: Detects PetitPotam-style NTLM coercion and AD Certificate Services exploitation attempts observed in Teams AD hijacking campaigns
author: Security Arsenal Threat Intelligence
date: 2026/11/03
modified: 2026/11/03
tags:
    - attack.credential_access
    - attack.t1187
    - attack.t1557
    - attack.t1649
logsource:
    category: network_connection
    product: windows
detection:
    selection_rpc:
        Image|endswith: '\lsass.exe'
        DestinationPort: 445
        Initiated: 'true'
    selection_cert:
        Image|endswith:
            - '\certreq.exe'
            - '\certutil.exe'
        CommandLine|contains:
            - 'config'
            - 'template'
    selection_suspicious:
        Image|endswith:
            - '\python.exe'
            - '\bun.exe'
        CommandLine|contains:
            - 'petitpotam'
            - 'ntlmrelayx'
            - 'esc1'
            - 'esc8'
    condition: 1 of selection_*
falsepositives:
    - Administrative certificate management
    - Vulnerability scanning
level: critical
---
title: Quick Assist Execution Followed by Portable Runtime Deployment
id: 9c5d1e4f-0a6b-6c3d-1e5f-7a8b9c0d1e2f
status: experimental
description: Detects Quick Assist remote assistance sessions followed by execution of portable runtimes used in Teams AD hijacking campaigns
author: Security Arsenal Threat Intelligence
date: 2026/11/03
modified: 2026/11/03
tags:
    - attack.execution
    - attack.t1059
    - attack.t1218
    - attack.t1105
logsource:
    category: process_creation
    product: windows
detection:
    selection_quickassist:
        Image|endswith: '\quickassist.exe'
    selection_runtime:
        Image|endswith:
            - '\python.exe'
            - '\python3.exe'
            - '\bun.exe'
            - '\java.exe'
            - '\node.exe'
        ParentImage|endswith: '\quickassist.exe'
    selection_suspicious_path:
        Image|startswith:
            - 'C:\Users\Public\'
            - 'C:\Windows\Temp\'
            - 'C:\ProgramData\'
            - '%TEMP%'
            - '%APPDATA%'
    condition: selection_quickassist or (selection_runtime and selection_suspicious_path)
falsepositives:
    - Legitimate IT support using Quick Assist
    - Developer tools execution
level: high

KQL Hunt Query (Microsoft Sentinel)

KQL — Microsoft Sentinel / Defender
// LLMJacking and Credential Theft Hunt - Security Arsenal
// Run with 7-day lookback minimum
let LLMC2Domains = dynamic(["jet-night.com", "newapi.pro", "artex.nhhh.de", "direct.465415.xyz"]);
let PhishDomains = dynamic(["authgpt.info", "ai-billing.info", "url-y.uk"]);
let SuspiciousIPs = dynamic(["31.57.201.201", "38.246.114.118"]);
let SuspiciousHashes = dynamic([
    "06c5c14f043aee3bcc5843b5f50909623d5c9a6239a7c1fba9cddad4111469e7",
    "85b6c424117494d3056e88689dbc94c75e9afdb187a339714043aec0da109f17",
    "f326201269f1ee2f803ae435317507263bb6e0dbfdeb70bd1bb5446e2e46f001",
    "6c1cfc02b4d1e31bf9bbdb0693340c94",
    "7151dc47580618f3b4b3c2254ded341c",
    "e24244445fd3686fcd49e35a78fb921e",
    "291db418fb4a6d24c797dc870d5a1cec9a00c0bc",
    "6c4be1f10290e994d926f1270aaf34a6fd3c6ad3"
]);
union isfuzzy=true
    (
    DeviceNetworkEvents
    | where RemoteUrl has_any (LLMC2Domains) or RemoteUrl has_any (PhishDomains) or RemoteIP in (SuspiciousIPs)
    | project Timestamp, DeviceName, InitiatingProcessFileName, InitiatingProcessCommandLine, RemoteUrl, RemoteIP, RemotePort, ActionType
    | extend HuntType = "NetworkIOC"
    ),
    (
    DeviceProcessEvents
    | where SHA256 in (SuspiciousHashes) or MD5 in (SuspiciousHashes) or SHA1 in (SuspiciousHashes)
    | project Timestamp, DeviceName, FileName, ProcessCommandLine, SHA256, MD5, SHA1, InitiatingProcessFileName
    | extend HuntType = "FileHashIOC"
    ),
    (
    DeviceProcessEvents
    | where FileName in~ ("python.exe", "python3.exe", "bun.exe", "java.exe", "node.exe")
    | where ProcessCommandLine has_any ("petitpotam", "ntlmrelayx", "secretsdump", "dcsync", "esc1", "esc8")
       or ProcessCommandLine has_any ("rclone", "copy", "move") and ProcessCommandLine has_any ("config", "remote")
    | project Timestamp, DeviceName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine
    | extend HuntType = "ADAttackTooling"
    ),
    (
    DeviceProcessEvents
    | where InitiatingProcessFileName =~ "quickassist.exe"
    | project Timestamp, DeviceName, FileName, ProcessCommandLine, InitiatingProcessFileName, InitiatingProcessCommandLine
    | extend HuntType = "QuickAssistAbuse"
    ),
    (
    SecurityEvent
    | where EventID == 4768 or EventID == 4769
    | where TargetUserName contains "krbtgt" or ServiceName contains "krbtgt"
    | where IpAddress in (SuspiciousIPs) or WorkstationName has_any (LLMC2Domains)
    | project Timestamp, Computer, TargetUserName, ServiceName, IpAddress, WorkstationName
    | extend HuntType = "DCSyncIndicator"
    )
| sort by Timestamp desc

PowerShell IOC Hunt Script

PowerShell
#Requires -RunAsAdministrator
# Security Arsenal - LLMJacking & Credential Theft IOC Hunt
# Run on suspected compromised endpoints and domain controllers

param(
    [Parameter(Mandatory=$false)]
    [int]$DaysBack = 7,
    [Parameter(Mandatory=$false)]
    [switch]$ExportCSV
)

$ErrorActionPreference = 'SilentlyContinue'
$StartDate = (Get-Date).AddDays(-$DaysBack)
$Results = @()

Write-Host "[+] Security Arsenal IOC Hunt - $(Get-Date)" -ForegroundColor Cyan
Write-Host "[+] Hunting for LLMJacking, Phishing, and AD Compromise Indicators`n" -ForegroundColor Cyan

# Network IOCs
$SuspiciousDomains = @("jet-night.com", "newapi.pro", "authgpt.info", "ai-billing.info", "url-y.uk", "artex.nhhh.de", "direct.465415.xyz")
$SuspiciousIPs = @("31.57.201.201", "38.246.114.118")
$SuspiciousHashes = @(
    "06c5c14f043aee3bcc5843b5f50909623d5c9a6239a7c1fba9cddad4111469e7",
    "85b6c424117494d3056e88689dbc94c75e9afdb187a339714043aec0da109f17",
    "f326201269f1ee2f803ae435317507263bb6e0dbfdeb70bd1bb5446e2e46f001",
    "6c1cfc02b4d1e31bf9bbdb0693340c94",
    "7151dc47580618f3b4b3c2254ded341c",
    "e24244445fd3686fcd49e35a78fb921e"
)

# Check DNS Cache for malicious domains
Write-Host "[*] Checking DNS Cache..." -ForegroundColor Yellow
$DNSCache = Get-DnsClientCache | Where-Object { $_.Entry -match ($SuspiciousDomains -join '|') }
if ($DNSCache) {
    foreach ($Entry in $DNSCache) {
        $Results += [PSCustomObject]@{
            Type = "DNSCache"
            Indicator = $Entry.Entry
            Data = $Entry.Data
            Time = "N/A"
            Risk = "HIGH"
        }
        Write-Host "[!] DNS Cache Hit: $($Entry.Entry)" -ForegroundColor Red
    }
}

# Check active network connections
Write-Host "[*] Checking Active Network Connections..." -ForegroundColor Yellow
$Connections = Get-NetTCPConnection | Where-Object { $_.RemoteAddress -in $SuspiciousIPs -or ($_.RemoteAddress -ne "0.0.0.0" -and $_.RemoteAddress -ne "::" -and $_.State -eq "Established") }
foreach ($Conn in $Connections) {
    $Process = Get-Process -Id $Conn.OwningProcess
    $DNS = Resolve-DnsName -Name $Conn.RemoteAddress -ErrorAction SilentlyContinue
    if ($Conn.RemoteAddress -in $SuspiciousIPs -or $DNS.NameHost -match ($SuspiciousDomains -join '|')) {
        $Results += [PSCustomObject]@{
            Type = "NetworkConnection"
            Indicator = $Conn.RemoteAddress
            Data = "$($Process.ProcessName) (PID: $($Conn.OwningProcess)) -> $($Conn.RemoteAddress):$($Conn.RemotePort)"
            Time = "Active"
            Risk = "CRITICAL"
        }
        Write-Host "[!] Suspicious Connection: $($Process.ProcessName) -> $($Conn.RemoteAddress)" -ForegroundColor Red
    }
}

# Check file system for suspicious hashes and cloud credential access
Write-Host "[*] Checking File System..." -ForegroundColor Yellow
$SearchPaths = @("$env:TEMP", "$env:APPDATA", "$env:LOCALAPPDATA", "C:\Users\Public", "C:\Windows\Temp")
foreach ($Path in $SearchPaths) {
    if (Test-Path $Path) {
        Get-ChildItem -Path $Path -Recurse -File -ErrorAction SilentlyContinue | ForEach-Object {
            $Hash = Get-FileHash -Path $_.FullName -Algorithm SHA256 -ErrorAction SilentlyContinue
            if ($Hash.Hash -in $SuspiciousHashes) {
                $Results += [PSCustomObject]@{
                    Type = "MaliciousFile"
                    Indicator = $Hash.Hash
                    Data = $_.FullName
                    Time = $_.LastWriteTime
                    Risk = "CRITICAL"
                }
                Write-Host "[!] Malicious File Found: $($_.FullName)" -ForegroundColor Red
            }
        }
    }
}

# Check for cloud credential file access (LLMJacking indicator)
$CloudCredPaths = @(
    "$env:USERPROFILE\.aws\credentials",
    "$env:USERPROFILE\.azure\",
    "$env:USERPROFILE\.config\gcloud\",
    "$env:USERPROFILE\.kube\config"
)
foreach ($CredPath in $CloudCredPaths) {
    if (Test-Path $CredPath) {
        $RecentAccess = Get-ChildItem -Path $CredPath -Recurse -ErrorAction SilentlyContinue | Where-Object { $_.LastAccessTime -gt $StartDate }
        if ($RecentAccess) {
            foreach ($File in $RecentAccess) {
                $Results += [PSCustomObject]@{
                    Type = "CloudCredentialAccess"
                    Indicator = $File.FullName
                    Data = "Last Access: $($File.LastAccessTime)"
                    Time = $File.LastAccessTime
                    Risk = "HIGH"
                }
            }
        }
    }
}

# Check for Quick Assist execution and suspicious child processes
Write-Host "[*] Checking Quick Assist Execution..." -ForegroundColor Yellow
$QuickAssistEvents = Get-WinEvent -FilterHashtable @{LogName='Security'; Id=4688; StartTime=$StartDate} -ErrorAction SilentlyContinue | 
    Where-Object { $_.Message -match "quickassist.exe" }
foreach ($Event in $QuickAssistEvents) {
    $Results += [PSCustomObject]@{
        Type = "QuickAssistExecution"
        Indicator = "quickassist.exe"
        Data = $Event.Message.Substring(0, [Math]::Min(200, $Event.Message.Length))
        Time = $Event.TimeCreated
        Risk = "MEDIUM"
    }
}

# Check for portable runtime execution
$RuntimeEvents = Get-WinEvent -FilterHashtable @{LogName='Security'; Id=4688; StartTime=$StartDate} -ErrorAction SilentlyContinue | 
    Where-Object { $_.Message -match "python.exe|bun.exe|java.exe|node.exe" -and $_.Message -match "Temp|Public|AppData" }
foreach ($Event in $RuntimeEvents) {
    $Results += [PSCustomObject]@{
        Type = "PortableRuntime"
        Indicator = "python/bun/java/node"
        Data = $Event.Message.Substring(0, [Math]::Min(200, $Event.Message.Length))
        Time = $Event.TimeCreated
        Risk = "HIGH"
    }
}

# Check for certificate service abuse (ESC1/ESC8)
Write-Host "[*] Checking Certificate Service Events..." -ForegroundColor Yellow
$CertEvents = Get-WinEvent -FilterHashtable @{LogName='Security'; Id=4886,4887,4888; StartTime=$StartDate} -ErrorAction SilentlyContinue
foreach ($Event in $CertEvents) {
    if ($Event.Message -match "template" -and ($Event.Message -match "client authentication" -or $Event.Message -match "domain controller")) {
        $Results += [PSCustomObject]@{
            Type = "CertificateAbuse"
            Indicator = "ADCS"
            Data = $Event.Message.Substring(0, [Math]::Min(300, $Event.Message.Length))
            Time = $Event.TimeCreated
            Risk = "CRITICAL"
        }
    }
}

# Check scheduled tasks for persistence
Write-Host "[*] Checking Scheduled Tasks..." -ForegroundColor Yellow
$SuspiciousTasks = Get-ScheduledTask | Where-Object { $_.TaskPath -notmatch "Microsoft" -and ($_.Actions.Execute -match "python|bun|java|node|powershell" -or $_.Actions.Arguments -match "http|https") }
foreach ($Task in $SuspiciousTasks) {
    $Results += [PSCustomObject]@{
        Type = "ScheduledTask"
        Indicator = $Task.TaskName
        Data = "$($Task.Actions.Execute) $($Task.Actions.Arguments)"
        Time = "Persistent"
        Risk = "HIGH"
    }
}

# Summary and Export
Write-Host "`n[+] Hunt Complete. Findings: $($Results.Count)" -ForegroundColor Cyan
$Results | Sort-Object Risk, Time | Format-Table -AutoSize

if ($ExportCSV -and $Results.Count -gt 0) {
    $ExportPath = "$env:USERPROFILE\Desktop\IOC_Hunt_$(Get-Date -Format 'yyyyMMdd_HHmmss').csv"
    $Results | Export-Csv -Path $ExportPath -NoTypeInformation
    Write-Host "[+] Results exported to: $ExportPath" -ForegroundColor Green
}

if ($Results.Count -eq 0) {
    Write-Host "[+] No IOCs detected. Continue monitoring." -ForegroundColor Green
}

Response Priorities

Immediate (0-4 Hours)

Network Controls

  • Block all identified C2 domains and IPs at perimeter firewall, web proxy, and DNS resolver
  • Implement TLS inspection rules for jet-night.com and newapi.pro to identify encrypted C2 patterns
  • Add phishing domains to secure web gateway block categories
  • Alert on any historical connections to 31.57.201.201 and 38.246.114.118

Endpoint Controls

  • Deploy EDR block rules for all eight file hashes
  • Hunt for Quick Assist execution in the past 14 days across all endpoints
  • Isolate any systems with confirmed execution of portable runtimes from temporary directories
  • Review scheduled tasks on domain controllers and privileged access workstations

Identity Controls

  • Force password reset for any user with Quick Assist session in the past 14 days
  • Review Azure AD/AWS IAM for newly created service accounts or access keys
  • Audit AI API key usage for anomalies (sudden quota spikes, unusual geographic access)
  • Enable enhanced monitoring for AD CS certificate requests

24 Hours

Credential Rotation

  • Rotate all AI API keys (OpenAI, Anthropic, Azure OpenAI, AWS Bedrock) if cloud credential file access is detected
  • Rotate cloud service account credentials and enable workload identity federation where possible
  • Review and rotate any credentials stored in environment variables on affected AI developer platforms

Threat Hunting

  • Execute KQL hunt across all Sentinel workspaces
  • Run PowerShell hunt script on all domain controllers, PKI servers, and systems with AI development workloads
  • Review DNS logs for subdomain enumeration of blocked domains
  • Analyze email gateway logs for ChatGPT billing phishing messages

Vulnerability Management

  • Identify systems potentially vulnerable to CVE-2026-40217 through scanner correlation
  • Prioritize patching of internet-facing cloud management interfaces and AI platform integrations

1 Week

Architecture Hardening

  • Implement cloud billing alerts with $1,000 threshold and automated key disabling
  • Deploy conditional access policies requiring phishing-resistant MFA for AI platform access
  • Segment AI development environments from production Active Directory
  • Enable AD CS auditing and restrict certificate template enrollment to authorized systems
  • Disable Quick Assist or restrict to approved IT support accounts via AppLocker

Detection Tuning

  • Tune Sigma rules based on false positive analysis
  • Deploy custom analytics for LLMJacking billing anomaly detection
  • Implement UEBA rules for AI API usage patterns

Supply Chain Review

  • Inventory all AI developer platform integrations
  • Review third-party platform security assessments
  • Establish credential rotation procedures for platform-stored environment variables

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.