A human at a control panel still completed genuine Google and Okta workforce sign-ins after victims typed codes on brand-spoofed pages. According to Push Security on 20 November 2026, the GoodGuys AiTM phishing campaign has been live since at least mid-October 2026, using paid expert-interview and webinar lures that impersonate Accenture, Nvidia, SpaceX, Google Cloud, and other employers. Victims passed a captcha, landed on fake work Google or Okta screens, and submitted passwords plus SMS codes, authenticator codes, or push approvals that an operator relayed in real time to the real identity provider. The resulting session belonged to the attackers because the login still accepted transferable proof. Phishing-proof device-bound MFA stops that harvest before any session exists, because nothing useful remains for an operator to type or approve on the real origin.

What Push Security documented about GoodGuys logins

Push Security published the public teardown on 20 November 2026 after tracking infrastructure from mid-October. First captures on 13 October 2026 hit paid-interview-portal branding under Accenture and Spotify titles tied to a WPP Media frame, with backend traffic already aimed at attacker hosts. Okta screens appeared in builds by 15 October. Later skins covered Culver’s, American Express, Coca-Cola, Nvidia, SpaceX, and Google Cloud. Landing hosts rotated from AWS EC2 to Vultr around mid-November, and Nvidia or Google Cloud builds shifted backend traffic again late in the month.

Fake Google pages accepted only work accounts and rejected Gmail, Yahoo, and Outlook addresses. That filter marks enterprise identity theft, not a consumer mailbox sweep. The kit family also carried Okta username and password steps, Okta push prompts, and Okta Verify, authenticator, or SMS code screens. Luke Jennings at Push named the operation after a hidden label in every build: “GoodGuys sign-in simulation.”

GoodGuys does not sit as a reverse proxy on the real protocol. Victim inputs POST as JSON to an attacker backend while socket.io commands drive the fake UI like a live control panel. Push’s assessment is that an operator replays each captured value against real Google or Okta sign-in until the identity provider issues a normal work session. Related Coca-Cola phishing traffic in the same window used a modified Evilginx proxy; that is a different machine from the GoodGuys human-relay design. Adjacent Starship Product Briefing lure copy from early November reappears on the GoodGuys SpaceX skin even though kits and hosting differ, so inbox themes can feed more than one build.

Named compromised organizations, account totals, formal threat-group attribution beyond the GoodGuys moniker, and monetization paths past session theft are not in the public teardown. The technical path is still clear: transferable factors at workforce login handed the operator a finished session. For the full lure timeline and kit teardown, read the companion analysis on legacymfa.sucks.

Why every typed code and push still counted as valid proof

Initial access in this campaign was credential-phase work. Victims typed a password on the attacker page, then answered whatever second factor the kit requested. SMS codes, Google Authenticator codes, Google tap_code prompts, Okta push, and Okta Verify or SMS codes all moved through the same pipeline. Each of those values can leave the victim’s hands and arrive at someone else’s keyboard. The operator finished the real ceremony with that proof, so Google or Okta issued a legitimate session to the wrong party.

Factor on the fake page What the victim did What the operator did
Work password Typed it into attacker UI Replayed it on real IdP
SMS or authenticator code Entered the digits shown Typed the same digits live
Google tap or Okta push Approved a familiar prompt Benefited from the approval
Work-only email filter Used a corporate address Still harvested enterprise IDs

To the victim, the steps looked like a normal Google sign-in. Familiar logos, expected prompts, and a polished flow hide the fact that origin binding never happened. A push approval on a phone does not cryptographically bind to the real login host when the user is staring at an attacker page. An OTP is a short-lived shared secret anyone who sees it can spend once. That is why live relay beats “MFA was on” as a talking point. The second factor still traveled.

Session lifetime tweaks do not prevent the first successful relay. They only force the operator to run the lure again sooner. Detection after mail, drive, or SaaS access begins is cleanup for a login that already completed with phishable proof. Attackers who hold the issued session face no further authentication barrier until revoke or expiry. Prevention has to remove the transferable material before that first session exists.

Interview and webinar framing raised click trust without changing the identity math. Brand skins rotated weekly. Backend hostnames moved. The socket-driven UI stayed aimed at human replay of password plus MFA. Defenders who only hunt reverse-proxy fingerprints can miss a control-panel relay that still ends with attacker-held Google or Okta access.

How device-bound phishing-proof MFA removes the relay surface

Phishing-proof MFA built on device-bound public-key credentials changes the ceremony the operator needs. The private key never leaves hardware on the enrolled device. The identity provider stores only public keys. Authentication demands a fresh origin-bound signature for the real login host, not a code a stranger can read or a push a stranger can benefit from after coaching.

On a GoodGuys-style fake page, that signature cannot complete. The attacker origin is wrong. There is no OTP field to fill, no SMS channel to intercept, and no approval prompt that transfers value to a remote operator. The real Google or Okta service never sees valid proof from the enrolled key, so it never issues the work session the campaign is built to steal. That is prevention, not detection: the relay has nothing useful to carry.

Stage GoodGuys outcome with phishable MFA Outcome with device-bound phishing-proof MFA
Fake work login page Password and MFA captured No transferable secret to submit
Live operator replay Real IdP accepts relayed proof Real IdP never receives a valid signature
Session issuance Attackers hold a normal work token No session issues to the attacker
Later revoke work Needed after theft Avoided when login never completes

This disclosure does not document a helpdesk reset or attacker-enrolled authenticator after theft. The failure that is documented is the login itself. Closing transferable factors at that step is enough to stop this kit’s primary harvest path.

Same-device, hardware-bound credentials also remove the second-phone dependency that makes push fatigue and code readout natural. The device-bound architecture keeps the private material inside the TPM or secure element the user already trusts for work. An enterprise IdP that issues only hardware-bound signatures will not accept a typed substitute from a fake interview portal.

Malware on a machine that already holds a legitimate session can still steal cookies locally. That path is endpoint compromise, not remote operator relay of a login form. GoodGuys, as documented, never needed that bar. It needed victims to type or approve factors on the wrong origin. Remove those factors and the campaign’s cheap path closes.

Passkeys bind the ceremony to the real origin

Passkeys and FIDO2-style WebAuthn credentials place a key pair on the device and keep the private key in hardware. The signature the IdP verifies is origin-bound to the real login host. That property makes plain passkeys phishing-proof at authentication: a fake GoodGuys page is the wrong origin, so a correctly enforced ceremony will not produce a usable assertion for the attacker.

On a generic enterprise identity provider that accepts only hardware-bound signatures, the browser or OS authenticator signs a challenge tied to that provider’s origin. The private key never crosses the network as a reusable secret. Compared with SMS, authenticator apps, or push, there is nothing for an operator to copy into a parallel real-login tab.

Passkeys harden the act of signing in when the ceremony is enforced. For GoodGuys as disclosed, origin-bound login alone already breaks the live relay of typed MFA. This disclosure does not document a helpdesk reset or attacker-enrolled authenticator after theft.

Key takeaways for defenders

Workforce Google and Okta remain high-value targets for polished lures as long as login proof can be typed, read aloud, or approved out of band. Device-bound phishing-proof authentication ends that bargain by refusing to mint a session from transferable material in the first place.

FAQ

Would phishing-proof MFA have stopped the GoodGuys campaign at login?

Phishing-proof device-bound MFA would have stopped the GoodGuys campaign at login because operators depended on replaying passwords and MFA values victims typed or approved on fake Google and Okta pages. With no transferable secret and with signatures bound to the real origin and enrolled hardware, the real identity provider never receives proof it can accept, so the work session never issues to the attacker.

Why did SMS codes, authenticator apps, and push fail against GoodGuys?

SMS codes, authenticator apps, and push failed against GoodGuys because each factor is transferable: a digit string or approval can be captured on the attacker page and finished on the real Google or Okta service by a live operator. According to Push Security, the victim flow still looked like a normal work sign-in while socket-driven commands fed an operator control panel that replayed every value.

Do passkeys alone block operator-relayed fake Google pages?

Passkeys block operator-relayed fake Google pages at authentication when the ceremony is enforced, because the private key signs an origin-bound challenge for the real identity provider and will not satisfy a spoofed host. That is phishing-proof login when the ceremony is enforced. This disclosure does not document a helpdesk reset or attacker-enrolled authenticator after theft, so the documented failure is the login itself.

Can shorter Google or Okta session lifetimes stop this relay?

Shorter Google or Okta session lifetimes cannot stop the GoodGuys relay. Lifetime limits only shrink how long a stolen session remains useful after the identity provider has already accepted relayed password and MFA proof. Preventing the first successful login by removing transferable factors stops the token from being issued to the attacker at all.

How is GoodGuys different from Evilginx-style reverse-proxy AiTM?

GoodGuys is different from Evilginx-style reverse-proxy AiTM because Push Security states the core kit does not use a reverse proxy. Fake pages collect inputs and post them to an attacker backend while an operator replays each value against real Google or Okta. Related Coca-Cola lures in the same reporting window used a modified Evilginx proxy, which is a separate technique from the GoodGuys human-relay design.