Key facts
- In device code phishing, the victim authenticates at a genuine Microsoft page while the attacker captures the resulting authorization.
- Huntress found EvilTokens early in 2026 during an investigation of a phishing campaign that used device codes and abused the Railway Platform-as-a-Service platform.
- Kali365 seizes tokens via device code flows and reverse-proxy lures, then replays them into current Microsoft 365 sessions to keep access, and expands using built-in AI features.
- The attacker's device generates a device code, the victim enters it at microsoft.com/devicelogin, and the attacker's device receives OAuth access and refresh tokens.
Huntress researchers found EvilTokens early in 2026 and analyzed Knight Office and Kali365 alongside it, three phishing kits that capture and replay Microsoft 365 authentication tokens to hijack sessions even where organizations enforce multifactor authentication.sources
In device code phishing, the attacker’s device generates a code and the victim enters it at microsoft.com/devicelogin on a genuine Microsoft sign-in page. In this flow, the victim authenticates at a genuine Microsoft page while the attacker captures the resulting authorization. The attacker’s backend then receives OAuth access and refresh tokens. Replaying a stolen token places the attacker inside an active session where login and multifactor authentication are already complete.
The token carries no password and prompts no MFA challenge, so the session looks like a legitimate sign-in even though the attacker never entered credentials.sources
Huntress also identified Knight Office as an adversary-in-the-middle phishing kit, the second tradecraft technique identified by the researchers. In these attacks, the attacker relays the authentication flow in real time and captures session information after the victim completes multifactor authentication. Knight Office takes valid Microsoft 365 session tokens and sends them to an operator console, where an operator can replay them and maintain access across tenant accounts.sources
Kali365 bundles a full Phishing-as-a-Service operation into a single control panel. The FBI classified the platform as Phishing-as-a-Service, first observed in April 2026 and sold via Telegram starting at $250 for 30 days. The FBI public service announcement I-052126-PSA was published on May 21, 2026.sources
Hundreds of Kali365 attacks hit companies and agencies in April 2026 across manufacturing, education, government, and other sectors in North America and Europe. Every victim was using multifactor authentication. The kit exploits the OAuth device code flow, which renders password managers, credential monitoring, and multifactor enforcement irrelevant. Kali365 displays stolen tokens in the attacker’s dashboard, enabling persistent access to Outlook, Teams, and OneDrive.sources
Kali365 seizes tokens through device code flows and reverse-proxy lures, then replays them into current Microsoft 365 sessions to keep access. The kit also expands using built-in AI features. Its full feature set includes seizing tokens and maintaining access through replay, which Huntress and FBI-aligned reporting describe as a complete Phishing-as-a-Service offering.sources
EvilTokens surfaced early in 2026 during an investigation of a phishing campaign that used device codes and abused the Railway Platform-as-a-Service platform. Huntress followed more than 340 hacked organizations across five countries in one related campaign tied to the kit. A joint Microsoft Digital Crimes Unit and FBI effort led to the arrest of one alleged EvilTokens operator later in 2026.sources
A construction company employee received a message from a typosquatted domain pretending to be from a genuine senior vice president at a separate construction company. The message reached victims through a Google share link that hid the attacker’s domain. The fraudulent page mimicked a Teams scheduler, displayed the executive’s headshot, and proposed an introductory call about a long-term supply partnership.sources
Selecting Continue triggered a Microsoft device code request on the attacker’s backend. The next page showed the code as a Booking Reference with a countdown timer. Choosing Copy or the confirm button put the code on the clipboard and opened a popup to Microsoft’s genuine sign-in page. Once the attacker’s backend confirmed a successful sign-in, the page closed the popup, showed a Verification complete notice, and redirected the victim to a Calendly booking page.sources
Attackers delivered the kit as AES-256-GCM ciphertext and included the decryption key on the same page, targeting e-mail scanners and crawlers that do not execute JavaScript. When Microsoft returned an error, the page told the victim it was preparing a new verification code, started a fresh device code session, and tried again. The attacker sent the final sign-in through a New Orleans residential proxy so the traffic looked like ordinary consumer traffic. Huntress’s security operations center later identified the valid session token the attacker obtained.sources

Successfully completed multifactor authentication does not by itself make a session trustworthy. Material Security co-founder and CEO Abhishek Agrawal said, “Once the attacker relays the sign-in and captures the session, they’re inside the account for as long as that session stays valid.” Huntress assesses that token theft is more valuable to attackers than password theft because it grants access to cloud resources and email.sources
Token-replay kits let attackers hijack already-authenticated Microsoft 365 sessions even where organizations enforce multifactor authentication. That makes traditional credential-based defenses insufficient without Conditional Access and session-protection controls.sources
The three kits illustrate a shift from credential theft to session theft, with device code phishing, adversary-in-the-middle relaying, and direct token replay as the core techniques. Credential monitoring and password managers cannot stop a valid session token issued after successful authentication.sources
Proofpoint documented a sharp increase in device code phishing beginning in September 2025, with state-aligned actors including Russia-linked groups first observed at scale. Financially motivated criminal actors adopted the technique by October 2025, and by February 2026 phishing-as-a-service tools like EvilTokens had commoditized it.sources
The FBI’s May 21, 2026 advisory on Kali365 and a Huntress security operations center compromise show the technique moving from research to active compromise. The observed victims included organizations that enforce multifactor authentication.sources
The NovaCookies adversary-in-the-middle service sells authenticated Microsoft 365 session theft for $320 a month or $200 for 14 days. Island researchers found the service used at least 755 domains, with more than half of targeted organizations in the US or related to US entities. NovaCookies infrastructure expanded sharply from mid-May through August 2026.sources
Microsoft Defender XDR issues two alerts for this behavior. The first is Suspicious Azure authentication through possible device code phishing, and the second is User account compromise via OAuth device code phishing.sources
Once defenders suspect a compromise, they should check mailbox activity, authentication methods, and new devices. Huntress’s SOC compromise checklist also includes unauthorized MFA methods, unrecognized registered devices, app consents that were not approved, unusual mailbox rules, and unexpected outbound e-mail. Attackers set inbox rules in April 2026 to move e-mails mentioning spam, phish, click, or link into concealed folders and mark them as read. They also registered a second device against some victim accounts to extend access beyond initial token expiry.sources
Containment should cover identity artifacts rather than only the original phishing lure. Defenders should revoke refresh tokens and remove unauthorized devices and Windows Hello credentials. Instead of relying only on password resets, organizations should revoke active sessions and refresh tokens after a suspected compromise. Teams should watch for replay or re-entry after containment and continue hunting once the original phishing domain disappears.sources
Teams should limit device code authentication when possible and examine mailbox access, forwarding rules, and outbound e-mail. The FBI and Arctic Wolf urge blocking device code flow in Microsoft Entra ID except when it serves a real operational need. Organizations should build a conditional access policy covering all users and cloud apps to block device code flow, allowing narrow exceptions for verified processes that require it. Defensive guidance across sources converges on blocking device code flow where not required, auditing existing usage, revoking refresh tokens and active sessions, and monitoring for anomalous token and mailbox activity.sources

Join the discussion