Cloud Infiltration: How a Fake IT Call Hands Attackers the Keys to Azure

Nobody picked a lock here. Someone phoned an employee, said "IT help desk, your passkey needs updating", and the employee opened the door.

This is a plain-English rewrite of what Microsoft's security team published about a cloud break-in campaign, as reported by The Hacker News on September 13, 2026. The accounts at stake are Microsoft cloud identities: Microsoft Entra ID sign-ins, the identity layer behind both Azure and Microsoft 365. If you run either, or just carry a work phone, it is worth five minutes.

The short version

Attackers phone or text employees on their personal phones, pretend to be IT (voice phishing), and talk them into "updating" a passkey or MFA on a fake Microsoft sign-in page. Once inside, they add their own second factor so they can return whenever they like. Then they quietly download everything the account can reach: files, mailboxes, the company directory.

Microsoft has seen this since May 2026.

Think of it as a building

  • Your Microsoft 365 account is an office in a big building.
  • Passwords and MFA are the lock.
  • The attacker does not pick it. They ring the tenant pretending to be maintenance and ask for the key.
  • Then they cut a copy (their own MFA device) so they never have to ask again.
  • Then they walk the halls (the Graph API), photocopy the cabinets (SharePoint, OneDrive, mail), and leave by a different door every time (rotating IP addresses).

The attack, step by step

  1. Research. They study the company from public sources like LinkedIn: who works where, who reports to whom.
  2. The call. A call or text to the employee's personal phone: "IT here, update your passkey, MFA or SSO now or lose access." A personal phone sits outside most company filters, which is likely why they use it.
  3. The fake sign-in page. The text links to a page that looks like Microsoft's. Two tricks are in play:
    • Adversary-in-the-middle: the fake page relays your login to the real one in real time and captures it.
    • Device code: you are walked through entering a code that quietly gives the attacker access. No password or cookie is stolen, and MFA never gets in the way, because you approved it yourself.
  4. Plant a second key. They enrol an MFA method they control: a new phone number, authenticator app or software token (modify authentication process: MFA, device registration). Now they can sign in without you, and keep doing so while your credentials or sessions stay valid.
  5. Walk the halls. Through the Microsoft Graph API, the same API legitimate apps use, they list users, groups, permissions and resources, and look at roles and high-value accounts for ways to climb higher.
  6. Take everything. High-volume downloads from SharePoint Online and OneDrive (data from cloud storage), mailbox contents (email collection), sometimes Exchange Online. It can run from hours to several days.
  7. Cover the tracks. Separate IP addresses for sign-in, recon and exfiltration, so one blocked address never connects the story.

A variant skips the call: they reuse credentials stolen in an earlier breach and simply register their own phone as an MFA method.

Why it is hard to catch

  • MFA was on, and it did not help. You handed over the key, or the attacker's own factor got registered.
  • Each Graph API call looks ordinary. Microsoft's warning: Graph abuse rarely looks suspicious through a single API call. The signal is the sequence.

The sequence is: odd sign-in, then a new MFA method, then a directory listing, then mass downloads. Any one of those alone can pass for a normal day.

The other campaign: same trick, different goal

In a separate run on August 3-5, 2026, attackers sent over a million emails posing as CEOs. They asked accounts-payable teams to wire ACH payments for a fake ServiceNow subscription (ServiceNow itself was not involved), backed by invented invoices, forged email threads and real executives' names in the signatures. The writing was AI-assisted, and the targets were US companies in IT services, consumer goods, real estate and manufacturing. Two of the fake domains: service-nowinc[.]com and domainlify[.]net.

That one steals money. The cloud campaign steals data. Both run on the same ingredient: a convincing story from a trusted-looking sender.

Who is behind it

Microsoft ties the initial access to several crews. Storm-3121 does the initial break-in that leads to ShinyHunters and Falcon extortion. Storm-3032, Microsoft's name for UNC6671, split from the BlackFile group and now works under the Helix extortion brand. The playbook overlaps with groups tracked as Cordial Spider, O-UNC-045 and PREY-0058.

The exact links between them are unclear. The likely story is shared playbooks and the same off-the-shelf phishing kits.

Their fake domains follow one pattern: the target's company name as a subdomain of a passkey-themed domain, so the link looks personal. Mandiant noted the same thing for UNC6671: credential-harvesting pages on generic passkey-themed root domains, with victim-specific subdomains for targeted voice phishing. Microsoft's examples of the root domains:

passkeyhelpdesk[.]com     secure-passkey[.]com
setupmypasskey[.]com      add-passkey[.]com
integratedsso[.]com       oktasession[.]com
syncmykey[.]com           portalsetuphub[.]com

The [.] is on purpose, so nobody clicks by accident.

What to do about it

For everyone

  • IT will not text your personal phone to "update your passkey". Hang up and call the help-desk number from your intranet.
  • Never type a code from a text or call into a Microsoft page (device code phishing).
  • Reached a sign-in page from a text message? Close it. Type the address yourself.

For admins (my takeaways; Microsoft's own advice includes the device-code and phishing-resistant MFA points)

  • Alert on every new MFA method registration (MITRE T1556.006), especially a phone number or authenticator app right after a sign-in from somewhere new.
  • Block device-code sign-in with Conditional Access if nobody needs it.
  • Prefer phishing-resistant MFA like passkeys and security keys: they are tied to the real site, so a look-alike page cannot replay them.
  • Watch bulk SharePoint and OneDrive downloads, especially from unmanaged devices, and correlate Graph API activity across events instead of call by call.
  • After an incident, remove unknown MFA methods and revoke sessions. A password reset alone does not evict them.

The whole post in four lines

  • Attackers phone staff, pose as IT, and walk them through a fake passkey or MFA update.
  • They add their own second factor, so a password reset alone does not kick them out.
  • Then they quietly read and download files and mail through the Graph API, from rotating IPs.
  • Catch the sequence, not the single event, and teach people that IT never texts a passkey link.

The mental model that stuck: the attacker's goal is not your password. It is a key they control.

Written in my own words; indicators are defanged.

References

The campaign

  • Microsoft Security Blog, Sep 9, 2026: Passkey-themed social engineering leads to identity and cloud compromise
  • Microsoft Security Blog, Sep 10, 2026: AI-assisted executive impersonation and invoice fraud (the CEO and ServiceNow campaign)
  • The Hacker News, Sep 13, 2026: Attackers Use Passkey Phishing to Hijack Microsoft Cloud Accounts and Exfiltrate Data
  • BleepingComputer: Passkey-themed phishing attacks lead to Microsoft 365 data theft

The crews

  • Google Cloud / Mandiant, Aug 2026: UNC6671 Rebrands: Multi-Brand Vishing Extortion Targets Financial Services and Enterprise Cloud Environments
  • The Hacker News, Aug 2026: UNC6671 Vishing Attacks Target Personal Phones to Steal SaaS Data
  • The Hacker News, Jan 2026: Mandiant Finds ShinyHunters-Style Vishing Attacks Stealing MFA to Breach SaaS Platforms

Background

  • Microsoft Security Blog, Jul 12, 2022: From cookie theft to BEC: Attackers use AiTM phishing sites as entry point to further financial fraud

Microsoft documentation

  • What is Microsoft Entra?
  • Microsoft Graph overview
  • OAuth 2.0 device authorization grant
  • Block authentication flows with Conditional Access policy
  • Conditional Access authentication strengths (phishing-resistant MFA)

MITRE ATT&CK

  • T1566.004 Phishing: Spearphishing Voice
  • T1556.006 Modify Authentication Process: Multi-Factor Authentication
  • T1098.005 Account Manipulation: Device Registration
  • T1530 Data from Cloud Storage
  • T1114 Email Collection