Threat reportPhishingTL-2026-6256
Modern Phishing Kits (EvilTokens, Knight Office, Kali365) Stealing Microsoft 365 Session Tokens via Device Code Phishing and AiTM
Three phishing-as-a-service kits (EvilTokens, Knight Office, Kali365) steal Microsoft 365 OAuth/session tokens through device code phishing and adversary-in-the-middle relay, then replay them into live sessions where login and MFA are already complete. Huntress documented all three in 2026; Microsoft disrupted EvilTokens in September 2026 and the FBI warned about Kali365 in May 2026.
Key facts
| Identifiers | TL-2026-6256 |
|---|---|
| Also known as | EvilTokens, Knight Office, Kali365, Octopi365, Freedom365, OctoLink Live |
| Severity | HIGH |
| Affected products | Microsoft Microsoft 365 / Entra ID (OAuth device code flow, Authentication Broker, Exchange Online) |
| Attribution | Storm-2992, Knight Office, Kali365 operators unattributed (medium confidence) |
| Malware | OctoLink Live, OctoLink Sender, telegram |
| Tools | EvilTokens, Kali365, Knight Office |
| MITRE ATT&CK techniques | T1078.004 Valid Accounts: Cloud Accounts, T1090.003 Proxy: Multi-hop Proxy, T1098 Account Manipulation, T1098.005 Account Manipulation: Device Registration, T1111 Multi-Factor Authentication Interception, T1114.002 Email Collection: Remote Email Collection, T1204.001 User Execution: Malicious Link, T1528 Steal Application Access Token, T1534 Internal Spearphishing, T1550.001 Use Alternate Authentication Material: Application Access Token, T1557 Adversary-in-the-Middle, T1566.002 Phishing: Spearphishing Link and 5 more in the attack flow |
| Motivation | FINANCIAL |
| Target sectors | construction, finance, real estate, health, legal, government administration, nonprofit, manufacturing, wholesale distribution, higher education |
| Target regions | North America, Europe, 053 - Australia and New Zealand, india |
| Detection rules | 9 (1 shown free) |
| Sources | 13 cited sources |
Detection coverage
As of 2026-11-17, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-6256 across Splunk SPL, Microsoft KQL and Sigma, covering 23 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: EvilTokens / Kali365 Device Code Sign-In From Kit Infrastructure or Scripted Client — Exploitation
Detects successful Entra ID (Microsoft 365) sign-ins that used the OAuth device code flow (T1528, T1078.004) and originate from infrastructure tied to TL-2026-6256: Railway egress ranges 162.220.232.0/22, 162.220.234.0/22 and 152.55.176.0/20 (EvilTokens), Tencent Cloud AS132203 43.173.64.0/20 (Kali365), the Knight Office console host 104.37.188.94, or a python-requests / OctoLink / BAV2ROPC user agent. Device code phishing completes login and MFA in the victim's own session, so the first signal is the token redemption sign-in.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1528, T1078.004, T1111, T1566.002, T1204.001
- Data sources: Entra ID sign-in logs (azure:monitor:aad SignInLogs/NonInteractiveUserSignInLogs), SigninLogs, AADNonInteractiveUserSignInLogs
- Severity: high
- Confidence: high
Splunk SPL
index=azure sourcetype="azure:monitor:aad" category IN ("SignInLogs","NonInteractiveUserSignInLogs") earliest=-24h properties.status.errorCode=0 (properties.originalTransferMethod="deviceCodeFlow" OR properties.authenticationProtocol="deviceCode")
```Successful device code sign-ins - EvilTokens, Kali365 and Knight Office token capture```
| eval user=lower('properties.userPrincipalName'), src_ip='properties.ipAddress', ua='properties.userAgent', app='properties.appDisplayName'
| eval kit_range=if(cidrmatch("162.220.232.0/22",src_ip) OR cidrmatch("162.220.234.0/22",src_ip) OR cidrmatch("152.55.176.0/20",src_ip) OR cidrmatch("43.173.64.0/20",src_ip) OR src_ip="104.37.188.94",1,0)
| eval script_ua=if(match(ua,"(?i)(python-requests|octolink-live|octolink-sender|BAV2ROPC)"),1,0)
| where kit_range=1 OR script_ua=1
| stats count AS events, dc(src_ip) AS source_ips, values(src_ip) AS src_ip_list, values(ua) AS user_agents, values(app) AS apps, max(kit_range) AS kit_range, max(script_ua) AS script_ua, min(_time) AS first_seen, max(_time) AS last_seen BY user
| eval risk=case(kit_range=1 AND script_ua=1,"critical",kit_range=1,"high",true(),"medium")
| convert ctime(first_seen) ctime(last_seen)
| table user risk events source_ips src_ip_list user_agents apps first_seen last_seenMicrosoft KQL
let KitRanges = dynamic(["162.220.232.0/22","162.220.234.0/22","152.55.176.0/20","43.173.64.0/20","104.37.188.94/32"]);
union isfuzzy=true SigninLogs, AADNonInteractiveUserSignInLogs
| where TimeGenerated > ago(24h)
| where ResultType == "0"
| where OriginalTransferMethod =~ "deviceCodeFlow" or AuthenticationProtocol =~ "deviceCode"
| extend KitRange = ipv4_is_in_any_range(IPAddress, KitRanges), ScriptUA = UserAgent matches regex "(?i)(python-requests|octolink-live|octolink-sender|BAV2ROPC)"
| where KitRange or ScriptUA
| summarize Events = count(), SourceIPs = make_set(IPAddress, 20), UserAgents = make_set(UserAgent, 10), Apps = make_set(AppDisplayName, 10), KitHits = countif(KitRange), ScriptHits = countif(ScriptUA), FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated) by UserPrincipalName
| extend Risk = case(KitHits > 0 and ScriptHits > 0, "critical", KitHits > 0, "high", "medium")
| project UserPrincipalName, Risk, Events, SourceIPs, UserAgents, Apps, FirstSeen, LastSeenSigma
title: Successful Device Code Sign-In From Phishing Kit Infrastructure or Scripted Client
id: 3b8f4c1e-7d52-4a96-b0e3-1f9a6c2d8e47
status: experimental
description: Detects successful Entra ID sign-ins using the OAuth device code flow from ranges tied to EvilTokens (Railway egress), Kali365 (Tencent Cloud AS132203) and the Knight Office console host, or from a python-requests or OctoLink user agent.
references:
- https://www.huntress.com/blog/railway-paas-m365-token-replay-campaign
- https://www.huntress.com/blog/kali365-device-code-phishing-kit
- https://www.ic3.gov/PSA/2026/PSA260521
author: Threadlinqs Team
date: 2026-11-17
tags:
- attack.credential_access
- attack.t1528
- attack.initial_access
- attack.t1078.004
- attack.t1111
- attack.t1566.002
- attack.execution
- attack.t1204.001
logsource:
product: azure
service: signinlogs
detection:
selection_flow:
ResultType: 0
OriginalTransferMethod: deviceCodeFlow
selection_infra:
IPAddress|cidr:
- 162.220.232.0/22
- 162.220.234.0/22
- 152.55.176.0/20
- 43.173.64.0/20
- 104.37.188.94/32
selection_ua:
UserAgent|contains:
- python-requests
- octolink-live
- octolink-sender
- BAV2ROPC
condition: selection_flow and (selection_infra or selection_ua)
falsepositives:
- Legitimate device code flow for headless devices, CLI tools or IoT from an approved hosting provider
- Internal automation using python-requests with device code authentication
level: highKnown false positives:
- Legitimate device code authentication for CLI tools, Teams Rooms or headless devices sourced from an approved cloud hosting range
- Internal engineering automation that authenticates with a python-requests user agent
Tuning notes: Allowlist known device code users and approved egress ranges. Once device code flow usage is audited, drop the kit_range/script_ua filter and alert on every successful device code sign-in outside the allowlist. Refresh the CIDR list from new Huntress/Microsoft reporting because the EvilTokens lure domains were seized on 2026-09-22 but replay infrastructure moves.
- Knight Office Lure Email: Signature Request Subject, Lure Sender IP or Kit Domain Links — Exploitation · unlock with TL Creds
- Exchange Admin Abuse: Malicious Inbox Rules, Transport Rules, Connectors and DKIM Changes — Post-Exploitation · unlock with TL Creds
- Device Code Sign-In Followed by Non-Interactive Token Use From a New IP — C2 / Token Replay · unlock with TL Creds
- DNS or Network Access to Knight Office Phishing and Console Domains — Exploitation · unlock with TL Creds
- Windows Hello for Business / Device Registration Sign-In With Dsreg Agent From Kit Infrastructure — Post-Exploitation · unlock with TL Creds
- OctoLink Sender Bulk Mail From Compromised Mailbox — Lateral Phishing · unlock with TL Creds
- Post-MFA Authentication Broker Sign-In Followed by Entra Device Registration — Post-Exploitation Persistence · unlock with TL Creds
- Token Replay via BAV2ROPC, python-requests or OctoLink User Agents — C2 / Token Replay Infrastructure · unlock with TL Creds
Indicators of compromise
As of 2026-11-17, Threadlinqs Intelligence tracks 23 indicators of compromise for TL-2026-6256.
Sample indicators (3 of 23)
- IP address:
104.37.188[.]94 - Domain:
idoej[.]com - URL:
hxxps://login[.]microsoftonline[.]com/common/oauth2/v2.0/token
Defanged, TLP:CLEAR. Full set of 23 indicators: unlock with TL Creds.
Technical analysis
EvilTokens is a Telegram-sold PhaaS (advertised from mid-February 2026) that embeds a live Microsoft device code in a lure page and sends victims to the legitimate Microsoft device login page; once the victim approves, the operator receives OAuth access and refresh tokens. Huntress first documented the campaign on 2026-03-20 as a Railway PaaS token-replay operation (about 344-350 organizations compromised in roughly two weeks), with replay engines running from Railway egress ranges (162.220.232.0/22, 162.220.234.0/22 and others) and lures delivered through Cloudflare workers.dev, Vercel, AWS Amplify, security-vendor URL rewriters and compromised sites. Persistence was achieved with programmatic BAV2ROPC refresh twice a day. On 2026-09-22 Microsoft's Digital Crimes Unit (court authorization from the Eastern District of Virginia, with Cloudflare, Coinbase, OpenAI, Railway, SpyCloud, Shadowserver, TRM Labs and Health-ISAC) seized 50 websites and disabled 150+ additional domains; Microsoft tracks the operators as Storm-2992 and linked the service to 12,000+ compromised inboxes across 10,000+ organizations. UK Metropolitan Police arrested two men (aged 32 and 38) on 2026-09-11. The toolkit included an AI chatbot to mine victim inboxes for wire-fraud opportunities.
Knight Office (Huntress, published 2026-09-02) is an AiTM kit with a Flask-based operator console ('Login - KNIGHT OFFICE') that captures sessions and credentials, offers webmail access with auto-refresh and real-time visitor statistics. Lures are spoofed DocuSign-style signature requests, voicemail and SharePoint/Teams notices (self-spoofed sender headers, 'l' for 'i' typos) routed through Monday.com tracking links and a compromised Joomla site to .vu phishing domains protected by Cloudflare Turnstile. In an incident on 2026-08-18 the first post-MFA authentication came through the Microsoft Authentication Broker from a residential callback proxy, followed by rogue Entra device registration and a Windows Hello for Business key bound to the account, allowing passwordless re-entry after token revocation. Stolen tokens were replayed with python-requests from Tencent Cloud ranges and the console host 104.37.188.94.
Kali365 (also Octopi365/Freedom365 panels) was first seen in April 2026 and sold mainly via Telegram; the FBI IC3 issued PSA I-052126 on 2026-05-21. It drives device code flows (and an AiTM reverse proxy in its base edition), captures tokens to a Telegram bot, offers 33 lure templates, a domain marketplace on Cloudflare Workers, OxaPay crypto payments, and post-compromise modules: an AI-assisted BEC analyzer (code comments reference Claude Sonnet), credential scanner, Exchange admin abuse (rogue connectors, mail-flow rules, DKIM toggling) and desktop apps OctoLink Live (hidden-Chromium refresh-token exchange to obtain ESTSAUTH cookies) and OctoLink Sender (bulk phishing from victim mailboxes, 2,500 sends per token per day). Huntress observed device-code sign-ins from Tencent Cloud AS132203 (43.173.64.0/20) from 2026-05-18, peaking 2026-05-20.
No CVE is involved; the kits abuse legitimate OAuth device authorization and session mechanics. Defender guidance across sources: block or restrict device code flow via Conditional Access, block authentication transfer, hunt successful sign-ins from unfamiliar cloud/hosting ranges and BAV2ROPC/python-requests user agents, review new devices, authentication methods (including Windows Hello for Business keys) and mailbox rules, revoke refresh tokens then remove rogue devices and credentials, and keep hunting after lure domains disappear. Note the originating BleepingComputer article is Huntress sponsored content; the arrest was reported by secondary sources as a UK Metropolitan Police action rather than an FBI arrest.
MITRE ATT&CK techniques by tactic:
Initial Access
T1078.004 Valid Accounts: Cloud Accounts; T1566.002 Phishing: Spearphishing Link
Command and Control
T1090.003 Proxy: Multi-hop Proxy; T1665 Hide Infrastructure
Persistence
T1098 Account Manipulation; T1098.005 Account Manipulation: Device Registration
Credential Access
T1111 Multi-Factor Authentication Interception; T1528 Steal Application Access Token; T1557 Adversary-in-the-Middle
Collection
T1114.002 Email Collection: Remote Email Collection
Execution
T1204.001 User Execution: Malicious Link
Lateral Movement
lateral-movement
T1550.001 Use Alternate Authentication Material: Application Access Token
Resource Development
T1583.006 Acquire Infrastructure: Web Services; T1608.005 Stage Capabilities: Link Target
Defense Evasion
T1684.001 Impersonation; T1684.002 Social Engineering: Email Spoofing
Remediation
Immediate actions
- Revoke refresh tokens and active sessions for affected users, then remove unauthorized Entra devices, authentication methods and Windows Hello for Business credentials
- Review mailbox access, inbox/forwarding rules, mail connectors, transport rules and outbound phishing from affected accounts
- Block sign-ins from unused hosting ranges such as Railway CIDRs via Conditional Access Named Locations
- Hunt successful sign-ins with python-requests or BAV2ROPC user agents and from AS132203 / 43.173.64.0/20
Workarounds
- Create a Conditional Access policy blocking device code flow for all users with narrow documented exceptions (exclude emergency access accounts)
- Block authentication transfer so sessions cannot move from computers to mobile devices
- Audit current device code flow usage before enforcing the block
Longer-term hardening
- Require compliant or hybrid-joined devices and enable Continuous Access Evaluation
- Train users to recognize device code lures and unexpected code-entry prompts
- Monitor for token replay and re-entry after the original phishing domain is gone
Affected and fixed versions
- Microsoft — Microsoft 365 / Entra ID (OAuth device code flow, Authentication Broker, Exchange Online)
Vulnerable versions: Tenants permitting device code flow and token-only session trust
Fixed in: No patch; mitigated by Conditional Access blocking device code flow
Timeline
- EvilTokens makes its first public post on Telegram, after earlier advertising on the NOIRLEGACY GROUP forum; SpyCloud's earliest victim captures date to 2026-02-18.
- Huntress observes the first compromise replayed from Railway infrastructure; more cases follow on 2026-02-24.
- Huntress SOC detects a surge of token replay across dozens of organizations (344 organizations in the 2026-03-02 to 03-19 data).
- Huntress publishes the Railway PaaS Microsoft 365 token replay report; an update on 2026-03-23 attributes it to EvilTokens (about 350 compromises in two weeks).
- Kali365 first observed (April 2026, per FBI); Knight Office lures also first reported via Huntress SAT in April 2026.
- Huntress begins seeing device code sign-ins from Tencent Cloud AS132203, peaking on 2026-05-20 with 80+ successful logins.
- FBI IC3 issues PSA I-052126 warning about Kali365 device code phishing against Microsoft 365.
- Huntress publishes its Kali365 / Octopi365 analysis including the OctoLink desktop tools and AI BEC module.
- Huntress responds to an AiTM incident: post-MFA sign-in via Microsoft Authentication Broker from a callback proxy, rogue device registration and WHfB key binding.
- Huntress publishes Inside Knight Office, a New M365 AiTM Phishing Kit.
- UK Metropolitan Police arrest two men (aged 32 and 38) in connection with EvilTokens; some outlets report 2026-09-22.
- Microsoft DCU and partners announce seizure of 50 websites and disabling of 150+ domains tied to EvilTokens (Storm-2992), citing 12,000+ compromised inboxes.
- BleepingComputer publishes Huntress sponsored overview of the three kits.
Sources cited
- FBI IC3 PSA I-052126: Kali365 phishing-as-a-service targeting Microsoft 365 ic3.gov · Advisory · 2026-05-21
- Huntress: Device code phishing explained Huntress · Research
- Huntress: Unboxing phishing kits Huntress · Research
- Huntress 2026 Cyber Threat Report: State of Phishing Huntress · Research
- Inside Knight Office, a New M365 AiTM Phishing Kit (Huntress) Huntress · Blog · 2026-09-02
- Railway.com Abuse for Microsoft 365 Token Theft / EvilTokens (Huntress) Huntress · Blog · 2026-03-20
- Kali365 Device Code Phishing Kit (Huntress) Huntress · Blog · 2026-06-11
- Inside Modern Phishing Kits Built to Steal Microsoft 365 Sessions (Huntress, sponsored) BleepingComputer · News · 2026-11-17
- Microsoft takes down EvilTokens device code phishing service (The Hacker News) The Hacker News · News · 2026-09-22
- EvilTokens made phishing-as-a-service look easy. Then it got taken down (Security Affairs) securityaffairs.com · News · 2026-09-23
- Two arrested in UK after Microsoft EvilTokens takedown (The Record) The Record · News · 2026-09-23
- Microsoft disrupts AI platform behind 12,000 email breaches (Redmondmag) redmondmag.com · News · 2026-09-23
- Microsoft 365 users targeted by new phishing threat that bypasses MFA (Help Net Security, Kali365) Help Net Security · News · 2026-05-22
More from Storm-2992
Drafted by the Threadlinqs research pipeline from 13 sources; published ; last updated —.