Threat reportPhishingTL-2026-1492
Device Code Phishing Campaign Targets Microsoft 365 via OAuth Device Authorization Grant Abuse
ReversingLabs documents an active, unattributed phishing campaign that abuses Microsoft's legitimate OAuth 2.0 Device Authorization Grant flow to hijack Microsoft 365 / Entra ID accounts without stealing passwords. Business-themed lures direct victims to a ClickFix-style landing page that walks them through entering an attacker-supplied device code into the genuine Microsoft login flow, silently authorizing an attacker-controlled device and granting persistent account access.
Key facts
| Identifiers | TL-2026-1492 |
|---|---|
| Also known as | OAuth Device Code Phishing (ReversingLabs cluster), ClickFix-style Device Code Phishing |
| Severity | HIGH |
| Affected products | Microsoft Microsoft 365 / Entra ID (Azure AD) — OAuth 2.0 Device Authorization Grant flow |
| Tools | DeviceCode_Phishing_LandingPageHTML |
| MITRE ATT&CK techniques | T1027 Obfuscated Files or Information, T1036 Masquerading, T1071.001 Web Protocols, T1098.005 Device Registration, T1102 Web Service, T1102.002 Bidirectional Communication, T1204.001 Malicious Link, T1528 Steal Application Access Token, T1531 Account Access Removal, T1550.001 Application Access Token, T1566.001 Spearphishing Attachment, T1566.002 Spearphishing Link and 4 more in the attack flow |
| Motivation | UNKNOWN |
| Target sectors | technology, manufacturing, finance, health, professional-services |
| Target regions | North America, Europe |
| Detection rules | 9 (1 shown free) |
| Sources | 8 cited sources |
Detection coverage
As of 2026-07-18, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-1492 across Splunk SPL, Microsoft KQL and Sigma, covering 19 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: Device-Code Phishing Landing Page — Invisible Unicode Keyword Obfuscation
Detects the ClickFix-style device-code phishing landing page by matching HTTP response bodies containing Zero Width Space (U+200B), Word Joiner (U+2060), or Zero Width Non-Joiner (U+200C) characters embedded inside the keywords 'Agreement', 'Verify', or 'Microsoft', co-occurring with a device-code prompt string. Catches the exploitation/initial-access lure before the victim reaches the real Microsoft sign-in popup.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1566.001, T1204.001, T1027
- Data sources: proxy, web_proxy, email_gateway
- Severity: high
- Confidence: high
Splunk SPL
index=proxy sourcetype="web_proxy" (http_content_type="text/html" OR http_content_type="*html*")
| rex field=http_response_body "(?<obf_agreement>A\u200bg\u200breement|Ag\u2060ree\u2060ment)"
| rex field=http_response_body "(?<obf_verify>Ver\u200cify)"
| rex field=http_response_body "(?<obf_microsoft>Mi\u200bcro\u2060soft)"
| eval invisible_hit=if(isnotnull(obf_agreement) OR isnotnull(obf_verify) OR isnotnull(obf_microsoft),1,0)
| where invisible_hit=1 AND like(http_response_body,"%devicecode%")
| stats count min(_time) as first_seen max(_time) as last_seen values(dest_domain) as dest_domain values(src_ip) as src_ip by http_referrer
| where count>0
| table first_seen last_seen src_ip dest_domain http_referrer countMicrosoft KQL
// SPL is authoritative for this row's query column; see kql_query column below
DeviceProcessEvents | where Timestamp > ago(24h) | where 1==0Sigma
title: SPL Reference — Device Code Phishing Invisible Unicode
id: 6b0f4b1e-1492-4c9a-8a3e-tl149201aaaa
status: experimental
logsource:
category: proxy
product: web
detection:
selection:
c-uri|contains: 'devicecode'
condition: selection
level: highKnown false positives:
- Legitimate documentation/CMS pages that coincidentally contain zero-width Unicode formatting characters for accessibility or RTL/LTR text shaping
Tuning notes: Tune the regex keyword list to the organization's brand terms in addition to 'Agreement/Verify/Microsoft'; whitelist known internal CMS domains that legitimately embed zero-width joiners for typography.
- Device-Code Phishing HTML Attachment — Vendor-Estimate Lure Delivery · unlock with TL Creds
- Primary Refresh Token Issuance to Newly Registered Device via Device Code · unlock with TL Creds
- Two-Phase DNS Fingerprint — Phishing Host to Microsoft Auth Chain · unlock with TL Creds
- Device-Code Phishing Landing Page HTML — Combined Unicode + Bitshift Artifact · unlock with TL Creds
- Account Manipulation via OAuth Device Registration Abuse (Entra ID) · unlock with TL Creds
- Web Service C2 Beaconing — Device-Flow Polling Abused for Bidirectional Communication · unlock with TL Creds
- Entra ID Device-Code Sign-In From Unmanaged Device Followed by Device Registration · unlock with TL Creds
- Device-Flow Polling Beacon — Four-Second Interval dc= POST Burst · unlock with TL Creds
Indicators of compromise
As of 2026-07-18, Threadlinqs Intelligence tracks 19 indicators of compromise for TL-2026-1492.
Sample indicators (3 of 19)
- Domain:
corpexl[.]nl - Domain:
adhere[.]it[.]com - Domain:
horizonex[.]it[.]com
Defanged, TLP:CLEAR. Full set of 19 indicators: unlock with TL Creds.
Technical analysis
The campaign begins with phishing emails carrying HTML attachments styled as business documents (e.g., vendor estimates/quotes). The HTML references JPG lure imagery via Content-ID URLs wrapped in clickable anchor tags with algorithmically generated ID parameters, evading static content matching. Clicking through lands the victim on a ClickFix-style page requesting 'document review,' which displays a verification/device code and instructs the victim to copy it and sign in via Microsoft. The sign-in button opens a genuine Microsoft authentication popup (via window.open() to the legitimate aka.ms/devicelogin, Akamai-hosted) — the victim enters the phishing-supplied device code into Microsoft's own, real interface. On completion, the victim unknowingly authorizes the attacker's registered device against their account, granting the attacker OAuth access/refresh tokens without ever touching or transmitting the victim's password. This is a device-code variant of adversary-in-the-middle (AiTM) phishing: rather than proxying credentials through a fake portal, the attacker relays a legitimate device-authorization code and lets Microsoft's real infrastructure do the authentication, which lets the campaign sail past classic anti-phishing detections that look for spoofed login pages or credential-harvesting POSTs. The phishing kit's backend polls for device-code completion via POST requests roughly every four seconds using URL-safe base64 encoding, mirroring the standard OAuth device-flow polling interval used by legitimate CLI/IoT clients. Technical/evasion notables: the landing-page HTML embeds invisible Unicode formatting characters — Zero Width Space (U+200B), Word Joiner (U+2060), and Zero Width Non-Joiner (U+200C) — inside high-signal words such as 'Agreement,' 'Verify,' and 'Microsoft' to defeat keyword-based phishing detection. A bitshifted artifact string tied to Microsoft Entra ID's Security Token Service ('EvoStsArtifacts') appears in the page and requires a left-shift-by-6-bits transform before base64 decoding, an anti-analysis/obfuscation touch. Infrastructure abuses Cloudflare Workers (*.workers.dev subdomains) alongside a rotating set of freshly registered or compromised domains (300+ URLs catalogued by ReversingLabs) to host landing pages and evade domain-reputation blocking, while calling out to genuinely Microsoft-owned endpoints (aka.ms, login.microsoftonline.com, aadcdn.msftauth.net, login.live.com, browser.events.data.microsoft.com) for the real authentication leg — producing a DNS/network fingerprint that blends malicious and legitimate Microsoft traffic. No CVE or software vulnerability is involved; this is pure abuse of a legitimate, MFA-adjacent authentication flow designed for input-constrained devices (smart TVs, CLIs) that was never intended to be surfaced to end users via a phishing lure. The technique class is not new — Microsoft's Storm-2372 (a suspected Russia-aligned actor) has run device-code phishing against government, NGO, defense, telecom, and energy-sector targets since August 2024, evolving by February 2025 to register attacker devices and mint Primary Refresh Tokens for deeper persistence — and a broader wave of OAuth device-code phishing (340+ M365 organizations across five countries) was reported by industry researchers in March 2026, with Microsoft publishing a follow-up on an AI-enabled variant in April 2026. The ReversingLabs campaign analyzed here is a distinct, unattributed cluster sharing the same technique class, landing-page kit style (ClickFix), and infrastructure-abuse patterns (workers.dev) as those broader waves, but the source does not attribute it to Storm-2372 or any other named actor.
MITRE ATT&CK techniques by tactic:
Defense Evasion
T1027 Obfuscated Files or Information; T1036 Masquerading
Command and Control
T1071.001 Web Protocols; T1102.002 Bidirectional Communication
Persistence
command-and-control
Execution
Credential Access
T1528 Steal Application Access Token; T1621 Multi-Factor Authentication Request Generation
Impact
lateral-movement
T1550.001 Application Access Token
Initial Access
T1566.001 Spearphishing Attachment; T1566.002 Spearphishing Link
Resource Development
T1583.006 Web Services; T1584.001 Domains
reconnaissance
Remediation
Immediate actions
- Block or restrict the OAuth 2.0 Device Authorization Grant flow in Microsoft Entra ID Conditional Access policies for users/roles that do not require it
- Deploy the ReversingLabs YARA rule (DeviceCode_Phishing_LandingPageHTML) at email/web gateways to detect landing pages containing invisible Unicode formatting characters near device-code/verification keywords
- Block known campaign domains and workers.dev subdomains identified in the ReversingLabs IOC list at web/email proxy
- Alert on Entra ID sign-in logs where authentication method = device code, especially from unmanaged/unregistered devices
Workarounds
- Disable the device code authentication flow tenant-wide via Conditional Access if no legitimate business use case exists
Longer-term hardening
- Where device code flow is operationally required (CLI/IoT), scope it via Conditional Access to specific named applications, locations, and compliant/managed devices only
- Deploy network detection for the identified two-phase DNS sequence (phishing host -> aka.ms -> login.microsoftonline.com -> aadcdn.msftauth.net -> login.live.com, followed by the post-code-entry sequence) combined with ~4-second interval beacon POSTs
- Enable Continuous Access Evaluation (CAE) and token binding where available to reduce the value of stolen device-flow tokens
- User-awareness training specifically covering device-code prompts and 'enter this code on another device' language, distinct from generic phishing-link training
Affected and fixed versions
- Microsoft — Microsoft 365 / Entra ID (Azure AD) — OAuth 2.0 Device Authorization Grant flow
Vulnerable versions: Any tenant with device code authentication flow enabled
Fixed in: Tenants with device code flow blocked or scoped via Conditional Access
Timeline
- Microsoft Threat Intelligence first tracks Storm-2372 running device-code phishing against government, NGO, defense, telecom, and energy-sector targets — establishing the technique class this campaign belongs to.
- Microsoft publishes analysis of Storm-2372's shift (observed Feb 14, 2025) to registering attacker-controlled devices via the Authentication Broker client ID and minting Primary Refresh Tokens for deeper persistence.
- Sophos publishes 'OAuth's Device Code Flow Abused in Phishing Attacks,' corroborating the technique class and reinforcing detection guidance ahead of the ReversingLabs disclosure.
- Proofpoint publishes 'Access granted: phishing with device code authorization for account takeover,' documenting device-code phishing as an emerging account-takeover vector against enterprise Microsoft 365 tenants.
- Industry researchers (Cloud Security Alliance / The Hacker News) report a broader OAuth device-code phishing wave hitting 340+ Microsoft 365 organizations across five countries.
- Microsoft Security publishes follow-up analysis of an AI-enabled device-code phishing campaign, reflecting continued evolution of the technique class.
- The ReversingLabs-documented campaign is observed actively delivering business-themed (vendor estimate) phishing lures leading to a ClickFix-style device-code landing page, ahead of public disclosure.
- ReversingLabs publishes full technical analysis of the campaign, including 300+ malicious URL IOCs, the Unicode-obfuscation and bitshifted-string techniques, and a detection YARA rule (DeviceCode_Phishing_LandingPageHTML).
- Campaign ingested into the Threadlinqs Intelligence platform for detection-engineering coverage (TL-2026-1492).
Sources cited
- Storm-2372 conducts device code phishing campaign Microsoft · Vendor report · 2025-02-13
- OAuth Device Code Phishing Hits 340+ Microsoft 365 Organizations labs.cloudsecurityalliance.org · Research · 2026-03-25
- Access granted: phishing with device code authorization for account takeover Proofpoint · Vendor report · 2026-01-01
- Inside an AI-enabled device code phishing campaign Microsoft · Vendor report · 2026-04-06
- OAuth's Device Code Flow Abused in Phishing Attacks Sophos · Vendor report · 2026-01-01
- Steal Application Access Token, Technique T1528 - MITRE ATT&CK MITRE · Research · 2026-01-01
- Device Code Phishing Campaign reversinglabs.com · Blog · 2026-06-12
- Device Code Phishing Hits 340+ Microsoft 365 Orgs Across Five Countries via OAuth Abuse The Hacker News · News · 2026-03-25
Previous and next in phishing
- Massive Smishing Campaign Abuses Gemini AI to Target Mobile Users with Fake Toll and Delivery Texts (Outsider Enterprise / Google v. Does 1-25) Previous report
- Coordinated Domain Impersonation Campaign Exploits Fable 5/Mythos 5 AI Model Export-Control Ban — 117+ Malicious Domains Targeting Anthropic, Claude, and Fable Brands Next report
Drafted by the Threadlinqs research pipeline from 8 sources; published ; last updated —.