---
title: Phish Wire - Sept 23, 2026
description: Fourteen phishing campaigns hit FL, KY, MN, NV, and TX orgs via cloud-hosted Microsoft scareware, fake CAPTCHA gates, and Evite and Instagram lures harvesting logins and OTPs.
---

[Phish Wire](https://blog.identityautomation.com/phish-wire)

# [Phish Wire - Sept 23, 2026](https://blog.identityautomation.com/phish-wire/phish-wire-sept-23-2026)

 Written by [Identity Automation](https://blog.identityautomation.com/phish-wire/author/identity-automation) | Oct 7, 2026, 7:48:39 PM

Between September 10, 2026 and September 23, 2026, our team tracked fourteen distinct phishing incidents across organizations in Florida, Kentucky, Minnesota, Nevada, and Texas, spanning tech-support scams, credential harvesters, and multi-stage OTP capture kits. Credential capture mechanisms varied widely: plain form-POSTs to attacker-owned PHP endpoints, double-submit password validation tricks, fake CAPTCHA gates stalling automated review, WebSocket and Supabase-backed submission channels, and direct exfiltration via Telegram Bot API. Impersonated brands centered heavily on Microsoft, appearing as fake 365 sign-ins, Defender lockscreens, and generic helpdesk lures, with secondary targeting of Instagram, Yahoo, AOL, Gmail, and Evite and Greenvelope invitation branding.

Infrastructure choices leaned on legitimate cloud platforms for reputation cover, with Azure Front Door, Azure Static Web Apps, and Bunny CDN each hosting multiple lures alongside disposable domains on .pw, .top, and ct.ws. What stands out this period is how many pages defer their actual harvesting logic into external JS bundles, third-party backends, or proxied Google resources, leaving static HTML review and markup-based scanning unable to surface the real capture mechanism. Several kits also show layered validation, from fake incorrect-password loops to Authenticator push simulation, indicating a shift toward tooling built to survive a second glance rather than a single convincing screen.

## Domains Reviewed

- pxus3ddy577dkeqhes23tnp2.b-cdn\[.\]net/ (4 variants)
- dcdrgtrdf-hub7d9g4d7agfcc8.z01.azurefd\[.\]net/wcdbcd/index.html?bcda=1-85...
- table4.cipit33\[.\]org/window2\_table9\_deer5\_ship3\_window4\_sugar9\_morning1\_...
- login.jumbia\[.\]com/?45ac78=I\_q0zArvDB8YqhnrVPCMBxHdxM36xq3zhQ
- check.zerix\[.\]vu/p/invitation/
- arrangement\[.\]top/checkkGuyII/DynaView/login.microsoftonline/pass.php?id...
- virwear\[.\]es/join\_us/rssv/
- psaartifacts\[.\]store/Wpwmwnll8fVsRl8dDj8MvUJgjxlmfCuE/orr?d85E...
- serumos\[.\]best/ (3 variants)
- votelivecontest-ml.ct\[.\]ws/en/1
- viruswarning0915us29yybx.z13.web.core.windows\[.\]net/
- hummmbers-c2ekaqh2gaergxhf.z03.azurefd\[.\]net/wcdbcd/index.html?bcda=1-88...
- regisfhnrl\[.\]pw/invite/Login/Ecard/
- truffledbull.wrarc\[.\]ca/search.html

A fake Windows Helpdesk support page targeted a Florida and Texas organization, using a tech-support impersonation lure rather than a brand login clone. Detections occurred on September 10, 2026, September 13, 2026, September 18, 2026, and September 23, 2026, indicating a sustained or repeating campaign rather than a one-off blast. The page's core mechanic is a scripted fake CAPTCHA overlay that gates the actual application root, hiding the credential-collection logic (bundled into an external index-DvsLmbh3.js file rather than inline markup) until a visitor clicks through a staged reCAPTCHA-style checkbox, a pattern built to defeat automated crawlers and quick manual review alike. The page is hosted on bunny.net's CDN infrastructure (b-cdn\[.\]net), a legitimate content delivery service that lends the URL a neutral reputation and lets it slip past domain-reputation filters that would flag a freshly registered look-alike domain.

Related subdomain variants:

- boldlink7314.b-cdn\[.\]net
- fastsite9577.b-cdn\[.\]net
- pxus3ddy577dkeqhes23tnp2.b-cdn\[.\]net
- rapidvolt4592.b-cdn\[.\]net

A fake Windows Defender lockscreen posing as a Microsoft security alert targeted a Kentucky organization, steering victims toward a phone-based tech support scam rather than a traditional login form. Detection occurred on September 22, 2026. The page has no form action or fetch call to harvest credentials directly; the username and password fields are present but marked readonly, meaning the visible login box is decorative and the actual conversion point is the phone number 1-855-XXX-XXXX repeated five times across the page. The social-engineering load carries the technique: fabricated malware names ("system-spawn.dll," "fetching-system.dll"), a red "Access has been blocked" banner, and Run Scan/Decline buttons that go nowhere all exist to manufacture urgency and push the victim toward a phone call rather than a credential submission. Hosting on an Azure Front Door subdomain (z01.azurefd\[.\]net) with a randomized prefix is notable because it lets the actor ride Microsoft's own infrastructure reputation while paying nothing for a dedicated domain, and the noindex/nofollow meta tags show an attempt to keep the page out of search and crawler visibility.

A generic Microsoft-style sign-in page targeted a Kentucky organization, using a single-page credential lure with no visible brand-specific chrome beyond a "Sign in to your account" title and a modal-style overlay built for a fake account picker or error dialog. Detection was confined to September 21, 2026. The HTML shows no visible form action or fetch/XHR endpoint in the retrieved markup, but the presence of a socket.io client library loaded from a third-party CDN suggests the credential submission is handled over a WebSocket channel rather than a conventional POST, which would let the page relay entered credentials to a listener in real time and evade log-based detection that watches for classic form-POST traffic. The CSS names (hue-rotate filters, unset animations, a full-viewport modal overlay) point to a kit built around dynamically injected or cloned branding assets rather than a static template, meaning the visual identity may be swapped per target without altering the page logic underneath.

A credential-harvesting page impersonating Microsoft's converged sign-in portal targeted a Kentucky organization, using a spoofed domain (jumbia\[.\]com) built to mirror the real login.microsoftonline.com flow down to the ConvergedSignIn page ID, FIDO challenge fields, and desktop SSO configuration. Detection occurred on September 21, 2026. The page harvests credentials through the native Microsoft authentication flow itself, posting form data to urlPost at /WgfCFk/FFYhk on the attacker's own login.jumbia.com infrastructure rather than a bolted-on PHP script, which lets the kit inherit legitimate-looking error codes like sErrorCode 50058 and reuse Microsoft's own client-side validation and telemetry endpoints.

A fake Evite invitation page impersonating Microsoft, Yahoo, AOL, and Gmail login flows targeted a Minnesota organization, offering a menu of email providers to funnel victims into a credential form regardless of which they clicked. Activity was confined to September 19, 2026. Credentials are captured through a form POST to processmail.php, and the same page scripts a follow-on process.php submission that requests a six-digit verification code, meaning the kit is built to harvest OTP tokens in addition to passwords. The flow also fakes a failed login on first submission before advancing the victim to the OTP stage, a deliberate double-submit trick that both re-captures a possibly mistyped password and normalizes the idea that a second security step is expected.

A credential-harvesting page impersonating Microsoft 365 targeted a Nevada organization, using cloned Microsoft login branding served from aadcdn.msftauth\[.\]net assets to sit on top of attacker-controlled infrastructure at arrangement\[.\]top. Detections occurred on September 17, 2026. The page pre-fills the victim's email address in the identity banner before presenting a password field, a sequencing trick that skips the username step entirely and pushes the target straight into a password-only prompt that feels like a continuation of a session they already started. Credential capture runs through a plain form POST to session/mlog.php on the same attacker domain, with the query string on the landing URL itself carrying what looks like a per-victim tracking token, suggesting individualized link generation rather than a single static page blasted to a list.

A credential-harvesting page impersonating a Greenvelope online-invitation notice, with provider-branded logins for Outlook, Office365, Yahoo, AOL, and generic email, targeted a Texas organization. Activity was confined to September 16, 2026. The kit runs a scripted multi-stage flow: the first form POSTs to processmail.php, then deliberately returns an "Incorrect Password" message to force a second submission, a common trick for validating that a victim typed a real password rather than a typo. After the second submission the page opens a fabricated OTP modal complete with a countdown timer, priming the victim to hand over a one-time passcode to process.php even though the actual AJAX call for that step is commented out in the source, meaning the OTP prompt exists purely as a social-engineering prop rather than a functioning capture mechanism at this stage.

A multi-stage Microsoft 365 credential-harvesting page targeted a Kentucky organization, complete with a full simulated login sequence covering password entry, OTP, SMS, phone-call approval, and Authenticator push. Activity was confined to September 16, 2026. The page pre-fills the victim's own address into the Microsoft sign-in template, a detail that makes the impersonation feel personalized before the target types anything. Ahead of that harvesting form sits a fake CAPTCHA gate that calls out to /captcha.php and /cverify.php for token issuance and verification, and only after a successful check does the page dynamically inject a remote script from user.cheacker\[.\]store to complete the credential flow, a pattern built to stall automated scanners and sandboxes that never solve the challenge.

A desktop-themed lure impersonating a Windows-style OS environment served as the front end for what appears to be a Microsoft account sign-in flow aimed at a Texas organization, using an "SerumOS" desktop shell to disguise a credential prompt as a familiar login experience. Detections occurred on September 11, 2026, September 14, 2026, and September 16, 2026. The visible HTML shows a fully custom desktop UI (windows, taskbar, start menu, dock) built to mimic a real operating system, with an embedded account module labeled with Microsoft's four-square logo and a step-based sign-in flow (ms-step, ms-field, ms-title) that walks a victim through identity entry before password capture, consistent with a staged credential-harvesting form rather than a single static page. The actual POST target and exfil logic sit in the referenced but untruncated core/auth.js and auth-vendor/supabase.js files, meaning the page leans on a third-party auth backend (Supabase) to process submitted credentials rather than a bespoke PHP handler, a pattern that lets the operator stand up a working login system quickly using legitimate infrastructure.

A credential-harvesting page impersonating Instagram targeted a Kentucky organization, disguised behind a voting-contest lure domain that has nothing to do with the brand it spoofs. Detection occurred on September 15, 2026. The page collects username and password through a local form POST to ./info/login.php, then independently exfiltrates the same credentials, along with victim IP and geolocation fields, via a hidden iframe pointed at a Telegram Bot API endpoint hardcoded into the page's JavaScript. A scripted fake-error loop rejects the first login attempt with a generic incorrect-password message and only forwards the victim after a second submission, a double-submit pattern that increases the odds of capturing a correct password on the retry and reduces victim suspicion.

A helpdesk-themed credential harvesting page titled "Helpdesk Windows Support" targeted a Kentucky organization, using a generic IT-support pretext rather than a specific brand to lower the victim's guard. Detection occurred on September 15, 2026. The page itself is a compiled single-page application, with all interactive logic bundled into an external index-DvsLmbh3.js module rather than inline HTML, meaning the actual credential-capture mechanism, whether a form POST or a fetch call to a collection endpoint, is deliberately opaque to static inspection of the markup and only recoverable by examining the JavaScript bundle directly. The bulk of the visible source is a sprawling CSS theming system supporting multiple branded product skins, which reads as reused infrastructure from a legitimate multi-tenant web application rather than a purpose-built phishing kit, suggesting the actors repurposed or cloned an existing SaaS front-end framework to host the lure.

A fake Windows Defender security alert page targeted a Nevada organization, pairing a spoofed Microsoft login box with a tech-support-scam call-to-action rather than a traditional credential form. Detection occurred on September 14, 2026. There is no form POST or fetch call in the HTML; the login fields are marked readonly and the Login and Cancel buttons point to href="#", meaning the page's actual function is to panic the viewer into dialing the listed hotline number, which appears seven times across the page including inside the fake alert text and the scrolling marquee footer. The visual social engineering does the work here: a locked-device message, a manufactured "Security Risk Detected: fetching-system.dll" line, and Windows/Microsoft branding combine to simulate an OS-level lockout rather than an ordinary phishing form.

A generic "Ecard invite" login page targeted a Minnesota organization, using a bare bones bootstrap login template with no brand-specific styling to speak of, just a background image and a modal form sitting on disposable .pw infrastructure at regisfhnrl\[.\]pw. Detection was confined to September 14, 2026. The visible HTML here is the front end only, a Bootstrap and jQuery shell with an image background and a modal login form, and the credential-capture endpoint itself never surfaces in the truncated markup, which is itself worth noting: the page's real work happens in script logic or a form action not present in what we captured, a pattern common to lightweight phishing kits that keep the harvesting mechanism in a separate file to survive takedown of the landing page alone.

A reCAPTCHA gating page sat in front of whatever payload was intended for a Texas organization, using Google's own enterprise reCAPTCHA branding and API to look like a routine bot check. Detection occurred on September 10, 2026. The page itself is entirely built around a reverse-proxy pattern: every external resource, from Roboto font files to the recaptcha\_\_en.js script, is rewritten through a local path (/browser/service/z6lkoeug/t0wkxr09/) that wraps the original gstatic\[.\]com and google\[.\]com URLs, meaning the attacker infrastructure is relaying live Google content rather than hosting a static clone. This is consistent with a proxy-based interstitial used to stall automated scanners and vet incoming traffic before a victim ever reaches a credential form, since no login fields or POST targets appear on this page at all.

## Recommendations

- Update phishing-detection logic to render and execute JavaScript rather than scan static HTML alone, since incidents on b-cdn\[.\]net, z13.web.core.windows\[.\]net, and psaartifacts\[.\]store all hid credential-capture logic in external JS bundles or remote-injected scripts that a markup-only scanner would miss.
- Extend network egress monitoring to flag outbound traffic to Telegram Bot API endpoints and hidden iframes carrying form data, the exact exfiltration channel used by votelivecontest-ml.ct\[.\]ws alongside its local login.php POST.
- Treat unfamiliar subdomains on trusted cloud platforms (azurefd\[.\]net, web.core.windows\[.\]net, b-cdn\[.\]net) with the same scrutiny as unknown domains by alerting on newly-seen randomized-prefix subdomains presenting login forms or OS-lockout banners, rather than allowlisting based on the parent platform's reputation.
- Build detection rules around the double-submit / fake-invalid-password pattern (seen in check.zerix\[.\]vu, virwear\[.\]es, and votelivecontest-ml.ct\[.\]ws) where a scripted "incorrect password" branch forces a second submission before an OTP modal appears — this behavioral sequence is a reliable phishing-kit signature independent of branding.
- Train users on the specific tech-support-scam pattern observed twice (z01.azurefd\[.\]net, z03.azurefd\[.\]net): a fake Windows Defender/lockscreen page with non-functional readonly login fields whose real goal is to get the victim to call a listed hotline number — reinforce that no legitimate Microsoft alert instructs users to phone a support line found on the page itself.
- Add URL-path pattern detection for phishing kits that mimic Microsoft path segments (e.g., arrangement\[.\]top's "login.microsoftonline/pass.php") or per-victim tracking tokens in query strings, since these durable naming conventions survive domain rotation better than content-based signatures.
- Brief users on the specific "already-in-session" trick seen in arrangement\[.\]top and psaartifacts\[.\]store, where the page pre-fills the victim's own email/address to simulate a continued login, and pair this with monitoring for CAPTCHA-gated pages that only load credential forms after a token exchange (psaartifacts\[.\]store, truffledbull.wrarc\[.\]ca), since this sequence is specifically built to stall sandboxes and automated crawlers.

Learn how PhishID can help protect your district from cutting-edge threats like these! [Schedule a Demo](https://www.identityautomation.com/request-a-demo)

[View full post](https://blog.identityautomation.com/phish-wire/phish-wire-sept-23-2026)

```json
{
  "@context" : "http://schema.org",
  "@type" : "BlogPosting",
  "author" : {
    "@type" : "Person",
    "name" : "Identity Automation"
  },
  "dateModified" : "2026-10-07T19:56:53.411Z",
  "datePublished" : "2026-10-07T19:48:39Z",
  "headline" : "Phish Wire - Sept 23, 2026",
  "image" : {
    "@type" : "ImageObject",
    "height" : 628,
    "url" : "https://www.identityautomation.com/hubfs/PW%209.23.png",
    "width" : 1200
  },
  "mainEntityOfPage" : "https://blog.identityautomation.com/phish-wire/phish-wire-sept-23-2026",
  "publisher" : {
    "@type" : "Organization",
    "logo" : {
      "@type" : "ImageObject",
      "height" : 60,
      "url" : "/hs/hsstatic/content_shared_assets/static-1.4092/img/default-amp-logo.png",
      "width" : 60
    },
    "name" : "Phish Wire"
  }
}
```