CISA Nominee, Connected Car Exploits, and AI Model Risks: 2026 Threat Landscape Analysis

---
title: OpenAI Codex API Unauthorized Access Attempt
id: 550d9a23-0f72-4e3a-a8c5-1d2f3b4c5d6e
status: experimental
description: Detects suspicious API calls to OpenAI Codex endpoints indicative of credential theft or token abuse.
references:
  - https://attack.mitre.org/techniques/T1552/
author: Security Arsenal
date: 2026/04/06
tags:
  - attack.credential_access
  - attack.t1552.001
logsource:
  category: proxy
  product: web
detection:
  selection:
    cs-host|contains:
      - 'api.openai.com'
    cs-uri|contains:
      - '/v1/engines'
      - '/v1/completions'
    cs-method: 'POST'
  condition: selection
falsepositives:
  - Legitimate development activity
level: medium
---
title: Microsoft Model Weight Repository Access
id: 7a3f1c82-9e4b-4d67-bc12-3e5a8f901234
status: experimental
description: Detects unauthorized access attempts to model weight storage or repositories.
references:
  - https://attack.mitre.org/techniques/T1530/
author: Security Arsenal
date: 2026/04/06
tags:
  - attack.collection
  - attack.t1530
logsource:
  category: file_event
  product: windows
detection:
  selection:
    TargetFilename|contains:
      - 'model.bin'
      - 'pytorch_model.bin'
      - 'tf_model.h5'
      - 'weights.pt'
    EventType: 'FileAccess'
  condition: selection
falsepositives:
  - Legitimate ML development
level: high
---
title: CAN Bus Injection via Diagnostic Port
id: 8b2f3c94-1a5d-4e78-9c23-4f6a7b8c9d0e
status: experimental
description: Detects suspicious diagnostic tool connections to vehicle networks.
references:
  - https://attack.mitre.org/techniques/T1547/
author: Security Arsenal
date: 2026/04/06
tags:
  - attack.persistence
  - attack.t1547
logsource:
  category: network_connection
  product: automotive
detection:
  selection:
    DestinationPort: 15765
    Protocol: 'ISO-TP'
  condition: selection
falsepositives:
  - Legitimate dealership diagnostics
level: medium

// Hunt for suspicious API access to AI services
DeviceNetworkEvents
| where TimeGenerated > ago(7d)
| where RemoteUrl contains "api.openai.com" or RemoteUrl contains "azure.microsoft.com"
| where InitiatingProcessName in ("powershell.exe", "cmd.exe", "python.exe", "curl.exe")
| project TimeGenerated, DeviceName, InitiatingProcessName, RemoteUrl, RemotePort, InitiatingProcessCommandLine
| order by TimeGenerated desc

// Hunt for model file access
DeviceFileEvents
| where TimeGenerated > ago(7d)
| where FileName contains "model.bin" or FileName contains "weights.pt" or FileName contains "pytorch_model.bin"
| where FolderPath contains "AppData" or FolderPath contains "Temp" or FolderPath contains "Downloads"
| project TimeGenerated, DeviceName, FileName, FolderPath, InitiatingProcessName, SHA256
| order by TimeGenerated desc

// Hunt for automotive diagnostic tool connections (via network telemetry)
DeviceNetworkEvents
| where TimeGenerated > ago(7d)
| where RemotePort == 15765 or RemotePort == 13400
| where InitiatingProcessName notin ("svchost.exe", "System")
| project TimeGenerated, DeviceName, InitiatingProcessName, RemoteIP, RemotePort
| order by TimeGenerated desc

-- Hunt for AI development tools accessing suspicious network endpoints
SELECT Pid, Name, CommandLine, Exe, Username, CreateTime, netstat() AS NetConnections
FROM pslist()
WHERE Name =~ '(python|powershell|cmd|curl)'
   AND CommandLine =~ 'openai|azure|api.openai.com'
   AND NetConnections.RemotePort =~ '443'

-- Hunt for model files in suspicious locations
SELECT FullPath, Size, Mtime, Ctime
FROM glob(globs='C:\\Users\\**\\AppData\\**\\model.bin')
WHERE Size > 1000000000

-- Hunt for automotive diagnostic tools
SELECT Pid, Name, CommandLine, Exe, Username
FROM pslist()
WHERE Name =~ '(canalyzer|canoe|vehicle|obd|uds|iso)'
   OR CommandLine =~ '15765|13400'

# AI Infrastructure Hardening Script
# Run on developer workstations and CI/CD servers

Write-Host "=== AI Infrastructure Security Hardening ===" -ForegroundColor Cyan

# 1. Audit OpenAI API keys and rotate if found in environment variables
Write-Host "[+] Checking for exposed OpenAI API keys..." -ForegroundColor Yellow
$envVars = Get-ChildItem Env: | Where-Object { $_.Value -match 'sk-[a-zA-Z0-9]{48}' }
if ($envVars) {
    Write-Host "[!] WARNING: OpenAI API keys detected in environment variables!" -ForegroundColor Red
    $envVars | ForEach-Object { Write-Host "  - $($_.Name)" -ForegroundColor Red }
    Write-Host "[!] ACTION REQUIRED: Rotate keys immediately and use Azure Key Vault or similar" -ForegroundColor Red
} else {
    Write-Host "[✓] No exposed API keys found in environment variables" -ForegroundColor Green
}

# 2. Audit local model repositories for unauthorized access
Write-Host "[+] Scanning for model weight files..." -ForegroundColor Yellow
$modelPaths = @(
    "$env:USERPROFILE\.cache\huggingface",
    "$env:USERPROFILE\.cache\torch",
    "C:\ProgramData\Microsoft\Azure\Models",
    "$env:USERPROFILE\AppData\Local\OpenAI"
)

foreach ($path in $modelPaths) {
    if (Test-Path $path) {
        $acl = Get-Acl $path
        $accessRules = $acl.Access | Where-Object { $_.IdentityReference -match 'Everyone|BUILTIN\Users' -and $_.FileSystemRights -match 'FullControl|Write' }
        if ($accessRules) {
            Write-Host "[!] WARNING: Overly permissive ACL on $path" -ForegroundColor Red
            $accessRules | ForEach-Object { Write-Host "  - $($_.IdentityReference): $($_.FileSystemRights)" -ForegroundColor Red }
            Write-Host "[!] ACTION: Restrict permissions to service accounts only" -ForegroundColor Red
        }
    }
}

# 3. Block suspicious outbound connections to AI APIs from non-dev processes
Write-Host "[+] Configuring Windows Firewall to restrict AI API access..." -ForegroundColor Yellow
$ruleName = "Block AI API from System Processes"
$existingRule = Get-NetFirewallRule -DisplayName $ruleName -ErrorAction SilentlyContinue

if (-not $existingRule) {
    New-NetFirewallRule -DisplayName $ruleName `n        -Direction Outbound `
        -RemoteAddress "api.openai.com","*.azure.com" `
        -Protocol TCP `
        -LocalPort 443 `
        -Action Block `
        -Program "C:\Windows\System32\svchost.exe","C:\Windows\System32\lsass.exe","C:\Windows\System32\csrss.exe" `
        -ErrorAction SilentlyContinue | Out-Null
    Write-Host "[✓] Firewall rule created: $ruleName" -ForegroundColor Green
} else {
    Write-Host "[✓] Firewall rule already exists" -ForegroundColor Green
}

# 4. Enable detailed logging for AI service access
Write-Host "[+] Enabling audit policy for API access..." -ForegroundColor Yellow
AuditPol /set /subcategory:"Filtering Platform Connection" /success:enable /failure:enable | Out-Null
AuditPol /set /subcategory:"Process Creation" /success:enable /failure:enable | Out-Null

# 5. Check for suspicious network connections to AI endpoints
Write-Host "[+] Checking active connections to AI services..." -ForegroundColor Yellow
$connections = Get-NetTCPConnection | Where-Object { 
    $_.RemoteAddress -match '13.107.246.|20.190.151.|40.126.32.' -and 
    $_.State -eq 'Established' -and 
    $_.OwningProcess -in (Get-Process | Where-Object { $_.Name -match 'svchost|lsass|csrss' }).Id
}

if ($connections) {
    Write-Host "[!] WARNING: Suspicious connections detected from system processes to AI endpoints!" -ForegroundColor Red
    $connections | ForEach-Object { 
        $proc = Get-Process -Id $_.OwningProcess -ErrorAction SilentlyContinue
        Write-Host "  - PID: $($_.OwningProcess) ($($proc.Name)) -> $($_.RemoteAddress):$($_.RemotePort)" -ForegroundColor Red
    }
    Write-Host "[!] ACTION: Investigate these processes immediately" -ForegroundColor Red
} else {
    Write-Host "[✓] No suspicious connections detected" -ForegroundColor Green
}

Write-Host "=== Hardening Complete ===" -ForegroundColor Cyan
Write-Host "Next Steps: Review logs in Event Viewer > Windows Logs > Security for Event ID 5156 (network connections)" -ForegroundColor Yellow