A cybersecurity team monitors cloud account security, device access, and threats across connected systems.
Phishing kits aimed at Microsoft 365 no longer just steal passwords. The current generation steals the session, which is the thing that matters once MFA is done. A Huntress Labs piece published on BleepingComputer on November 17 looks at three of these kits: EvilTokens, Knight Office and Kali365. That piece is sponsored content written by Huntress. The company sells security services, so treat its framing as vendor-authored. Microsoft and the FBI have separately documented much of the same ground, and this article draws on both.

A cybersecurity team monitors cloud account security, device access, and threats across connected systems. Why a finished MFA prompt proves little​

Huntress describes three techniques that turn a lure into a usable session:

  • Device code phishing. The attacker starts a real device-code sign-in. The victim is talked into entering the code on Microsoft's genuine page. Huntress says the victim never types credentials into a strange site, so it feels like a normal login.
  • Adversary-in-the-middle (AiTM). A reverse proxy relays the sign-in in real time. The victim completes MFA, and the attacker captures the resulting session information.
  • Token replay. The stolen token is used to get into services as an already-authenticated user.

The common thread is that MFA can be completed honestly and the session can still belong to the attacker. A completed prompt does not make the session trustworthy.

Microsoft's own write-up on device code phishing explains the design tradeoff. The flow exists for input-limited devices such as smart TVs, printers and Teams room devices. Because the user signs in on a separate device, the session that started the request is not strongly bound to the user's original context. That is the gap the attackers use.

Terminology matters here. In Microsoft's Entra documentation, access tokens issued by Entra ID last one hour by default. Refresh tokens let an app get new ones silently. Many browser apps use their own session cookies, and Entra cannot directly revoke those. Not every stolen artifact behaves the same way.

The three kits​

EvilTokens​

Huntress found EvilTokens in a device-code campaign that abused the Railway platform. Its March reporting counted more than 340 organizations across the US, Canada, Australia, New Zealand and Germany. Huntress was careful to say it saw abuse of Railway-hosted infrastructure but did not directly observe attackers using Railway's AI features.

Microsoft's September 22 reporting gives the larger picture:

  • The service appeared in February 2026. Microsoft links it to more than 12,000 compromised inboxes across over 10,000 organizations.
  • Microsoft says it seized 50 websites and disabled more than 150 further domains, working with partners.
  • Microsoft says the service was sold through Telegram for a $1,500 initiation fee and a $500 recurring subscription.
  • Microsoft describes an AI-style assistant that could analyze a compromised inbox. Preset prompts hunted for wire-transfer discussions, vendor invoices and the best people to impersonate.
  • The service used 44 lure themes. Microsoft reports redirects through compromised sites and serverless platforms such as Vercel, Cloudflare Workers and AWS Lambda.

On the legal side, Microsoft says the Metropolitan Police arrested two men on September 11 on suspicion of offenses connected with the alleged operation. Both were released on bail while the investigation continues. The Huntress article says one alleged operator was arrested, which is less precise than Microsoft's account. No one has been convicted, and the disruption does not mean every EvilTokens-style campaign has ended. Microsoft itself says the model will outlast the infrastructure.

Knight Office​

Knight Office is an AiTM kit. Huntress found its operator console while investigating an incident on August 18, 2026.

  • The lure: a DocuSign-style email, routed through a Monday.com tracking link and a compromised Joomla site.
  • The tell: the post-MFA sign-ins came from an IP flagged as a callback proxy. Huntress explains that in a legitimate login the sign-in and token-callback IPs line up. In AiTM the token issuance can complete through an attacker's proxy in an unrelated location.
  • Persistence: the attacker enrolled an unauthorized device in Entra ID and bound a Windows Hello for Business key credential. Huntress notes that this lets an attacker get back in after existing tokens are revoked.
  • Scale: Huntress telemetry tied at least nine phishing attacks to the kit's console IP over two weeks. That is Huntress's own visibility, not a count of all victims.

Huntress's detailed account of the lure pages has a wrinkle. The pages it recovered showed a device code, so the investigated chain mixed device-code elements with the AiTM framing. Anyone mapping this to a detection rule should read the raw details rather than the one-word label.

Kali365​

The FBI's May 21, 2026 public service announcement says Kali365 was first seen in April 2026 and was mainly distributed through Telegram. It offers AI-generated lures, campaign templates, tracking dashboards and OAuth token capture. Per the FBI, the attacker captures access and refresh tokens, then can reach Outlook, Teams and OneDrive without a password or further MFA challenge.

Huntress's later technical analysis, as summarized by a third-party digest, describes three variants and 33 lure templates. It also describes a token vault, an in-panel webmail proxy, and two Electron desktop apps that turn stolen tokens into browser sessions. Those details come from Huntress, not from independent verification. Don't assume every Kali365 deployment has every feature.

What defenders should do​

Hunt the whole chain​

Huntress's central advice is to follow the sequence, not a single alert. Look for:

  1. Successful sign-ins from unfamiliar cloud infrastructure.
  2. Device code authentication and automated token refresh.
  3. New devices and new authentication methods.
  4. Mailbox activity, forwarding rules and outbound phishing.

Block device code flow where you can​

Microsoft Learn documents a Conditional Access policy. The steps are:

  1. Sign in to the Entra admin center as at least a Conditional Access Administrator.
  2. Go to Entra ID > Conditional Access > Policies and select New policy.
  3. Under Users, include the users in scope (all users is recommended).
  4. Exclude emergency-access (break-glass) accounts, and audit that exclusion list regularly.
  5. Under Target resources, choose the apps in scope (All resources is recommended).
  6. Under Conditions > Authentication Flows, set Configure to Yes and select Device code flow.
  7. Under Grant, select Block access.
  8. Set the policy to Report-only first. After reviewing the impact, switch it to On.

Microsoft says to audit existing device-code use first and keep exceptions to documented, secured cases. Some Teams devices may need the flow. Microsoft advises scoping such an exception to specific Teams device resource accounts and excluding the Device Registration Service resource.

The FBI adds that admins can also consider blocking authentication transfer, which stops users moving a sign-in from a PC to a mobile device. Huntress suggests requiring compliant devices for Exchange Online and SharePoint as a medium-term hardening step.

Contain a suspected compromise​

Microsoft Learn's documented path in the Entra admin center is:

  1. Open the user and clear Account enabled.
  2. Select Revoke sessions on the user's Overview page.

For scripted or bulk response, the same documentation shows Microsoft Graph PowerShell. It uses Update-MgUser to disable the account, Revoke-MgUserSignInSession to revoke refresh tokens, and Update-MgDevice to disable registered devices. Role requirements apply. Cloud Device Administrator is the minimum for devices, and User Administrator or Privileged Authentication Administrator is needed for accounts.

Revocation has limits. Microsoft's threat intelligence team warns that standard revocation often only invalidates refresh tokens, so existing access tokens can stay live for up to an hour. It therefore recommends temporarily disabling the compromised account for immediate containment, despite the possible business disruption. Huntress's Railway write-up says the same about in-flight tokens unless continuous access evaluation (CAE) is enabled. App-issued session cookies need action by the app.

Revoking tokens is also not the end:

  • Remove unauthorized devices and any Windows Hello for Business credentials the attacker bound.
  • Inspect inbox rules, forwarding and outbound mail.
  • Keep monitoring for replay or re-entry after the phishing domain has gone dark.

Microsoft's Digital Crimes Unit adds a business-process point. Verify any request to change payment details or redirect funds through a trusted second channel. Mailbox access can turn into fraud within minutes.

The takeaway​

The kits differ in plumbing, but the lesson is the same. Identity telemetry is where these attacks show up, so alert on device-code sign-ins, unfamiliar infrastructure and surprise device registrations. For most tenants, blocking device code flow is the cheapest control, and it can be tested safely in Report-only mode first.

 

References

  1. Inside Modern Phishing Kits Built to Steal Microsoft 365 Sessions BleepingComputer 2026-11-17T10:01:11-05:00
  2. Inside Knight Office, a New M365 AiTM Phishing Kit | Huntress huntress.com
  3. Threat Actors Abuse Railway.com PaaS as Microsoft 365 Token Attack Infrastructure | Huntress huntress.com