Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

LevelBlue SpiderLabs screenshots of four administrative dashboards from the same phishing kit, branded @globaltoken1, SPY, BlackApex and OGPARKER1, each counting total sessions, captured PRT tokens and unique victims, with victim email addresses blurred.
Image: LevelBlue SpiderLabs (fair use). Figure 22 from the SpiderLabs report, published Oct. 30, 2026.

Detection

LevelBlue Finds More Than 70 Servers Running One Device Code Phishing Kit Under Names Like BlackApex and SPY, and Says the Code Bears the Marks of AI Assistance

The kit opens Microsoft's real device login page with the code already on the victim's clipboard, then holds live tokens and a Primary Refresh Token once the victim finishes MFA. SpiderLabs found the panels through a robots.txt hash and says Entra ID sign-in logs, not lure signatures, are where defenders will catch it.

Ava Okello, DetectionLondon6 min read

LONDON - A phishing kit that steals Microsoft 365 sessions through the device code sign-in flow is running on more than 70 servers under at least eight different panel names, and its code shows the hallmarks of AI-assisted development, researchers at LevelBlue SpiderLabs reported on Friday.

The kit does not match any publicly documented phishing kit, the researchers wrote. Its panels go by names including BlackApex, DarkApex, SPY, XXX101-TOKEN, OGPARKER1, W1-TOKEN, ALIBOBO-TOKEN and @globaltoken1, but the code, endpoints and interface are the same across all of them. SpiderLabs said it could not tell whether the names reflect separate operators or resellers and customers of a single kit, but the pattern "is consistent with a shared product being rebranded and redeployed."

The report, by Serhii Melnyk, Cris Tomboc and King Orande, adds a new family to a crowded field. Through 2026, ready-made phishing-as-a-service kits such as EvilTokens, Kali365 and Tycoon2FA have made device code phishing available to almost anyone willing to pay, SpiderLabs noted.

The victim completes MFA for the attacker

LevelBlue SpiderLabs screenshot of a SharePoint-themed lure page titled Verify to access documents, showing a verification code, a Copy Code button, three numbered steps to copy the code, click continue and paste, and sign in with a Microsoft account, and a Continue to SharePoint button.
Image: LevelBlue SpiderLabs (fair use).The SharePoint lure, part of Figure 6 in the report.

Every lure runs on the same engine. The page asks the attacker's backend for a device code, then checks in repeatedly to see whether the victim has finished signing in. On top of that sit at least five lures dressed as everyday business tasks, including SharePoint, Zoom, Adobe Acrobat Sign, Microsoft Excel and OneDrive. The Zoom lure copies Microsoft's "Permissions requested" screen and swaps in the victim's company logo and name based on their email domain.

In every variant, the main button opens Microsoft's real microsoft.com/devicelogin page in a pop-up, with the device code already placed on the victim's clipboard. Everything the victim sees on the Microsoft page is genuine. The attacker's server polls every five seconds, and when sign-in completes it redirects the victim to a real Microsoft Word page.

By then the server has redeemed the code and holds a live access token and refresh token for the account. "The victim completed MFA themselves, so the control meant to stop the attacker was satisfied by the wrong person," the researchers wrote. The kit also captures a Primary Refresh Token, which its panel displays and uses to reach the victim's mailbox. Because a PRT supports single sign-on across Entra ID integrated applications, SpiderLabs said, its theft can extend well beyond one session.

Both delivery chains SpiderLabs collected began with business-themed spearphishing and passed through a trusted cloud service: one through a Google share link, the other through a PDF that led to a shared Microsoft Forms page. Both ended at login.microsoftonline.com/common/oauth2/deviceauth, where the user is told they are signing in to "Microsoft Authentication Broker" on another device in the United States. That wording, the researchers wrote, is the giveaway.

Why SpiderLabs thinks AI wrote it

LevelBlue SpiderLabs screenshot of the kit's Continue button handler in JavaScript, with a comment saying the code is silently pre-copied to the clipboard so the victim just has to paste, followed by a window.open call to microsoft.com/devicelogin.
Image: LevelBlue SpiderLabs (fair use).Figure 9, the Continue button handler.

Hand-written phishing kits tend to be terse, often with no comments, the researchers wrote. This one is "suspiciously noisy," with emoticons, verbose console logs and structured documentation comments, and a "// Helper:" label above nearly every function. The scripts include defensive engineering that throwaway phishing pages rarely get: a flag that prevents duplicate polling timers, separate retry and restart paths, separate pop-up handling for mobile and desktop, and a timer that refreshes the device code after 14 minutes, just before Microsoft's own 15-minute expiry.

The strongest signal, in SpiderLabs' view, is consistency. All five lures share the same backend contract, the same getSessionId() pattern and the same retry logic while looking entirely different on the surface, which is what one would expect if one specification had been used to generate several skins. None of the code is obfuscated. Weighing the signals together, the researchers wrote that they "assess with high confidence that the codebase was produced largely through AI-assisted development."

LevelBlue SpiderLabs screenshot of the kit's JavaScript with comments above each step and emoji-marked console log messages such as Active landing path and Path mode, next to the kit's own log output showing a Turnstile check, the active landing path and a redirect to the loader.
Image: LevelBlue SpiderLabs (fair use).Figure 14, AI-assisted code annotations and verbose console logs.

How the panels were found

The operators left their own trail. The kit's robots.txt file listed /admin and /api paths and an administrative HTML file, pointing straight at the management portal. SpiderLabs took the SHA-256 hash of that robots.txt body, 5ec645bf49961d0f1b14e5ddc354e44df42b75b4451c19c6362e6a1d829a917f, and searched Censys for other servers returning the same file. That turned up more than 70 URLs hosting the same panel and lure, mostly on ports 5000 and 443 across several hosting providers, some associated with bulletproof hosting. Inside the panels, the researchers found multiple API paths reportedly accessible without authentication, exposing statistics, configuration, sessions, domains, anti-bot controls and access logs. The report ends with a table of VPS addresses and landing domains grouped by panel name.

What defenders can do

SpiderLabs argues that if AI-built kits become common, tracking named kit families will cover a shrinking share of the threat, because "a new front end and fresh domains can be produced in minutes." Its recommendations focus on the protocol rather than the lure:

  • Review Microsoft Entra ID sign-in logs for unusual deviceCodeFlow events, especially from unmanaged devices, unexpected IP ranges or unfamiliar user agents.
  • Where practical, block device code authentication with Conditional Access. Where legitimate tools depend on it, limit it to compliant devices, named locations or approved administrator groups.
  • Train users to treat any unexpected request to enter an access or device code as suspicious, even when the sign-in page is genuine Microsoft, and make suspicious messages easy to report.

For hunting, the robots.txt hash and the indicator table give teams a starting point for infrastructure pivots, with the researchers' caveat that hosting and domain resolution may change over time.

About the Author

Ava Okello

Ava Okello covers detection engineering, EDR telemetry, and SOC hunting for SOCtember from London.

Related stories

Detection desk