A free ChatGPT Pro or Grok seat on a fake event page is not a legitimate source of a corporate Microsoft 365 or Google Workspace session. According to Push Security on 5 November 2026, fake OpenAI, SpaceX, and Anthropic registration pages forced work-account sign-in, then split into Kali365 Microsoft device-code token theft and VoidProxy reverse-proxy capture of Google (and secondary Microsoft) passwords, live MFA responses, and session cookies. Device-bound phishing-proof authentication stops the VoidProxy credential relay so that reverse proxy never finishes a workforce login. Kali365 still issues tokens after a completed real Microsoft device grant, so defenders must also block unsafe device-code OAuth and revoke sessions rather than treat login cryptography as the whole fix.
What Push documented on the fake AI event RSVPs
Push researchers found fake registration pages for OpenAI, SpaceX, and Anthropic events that required corporate sign-in rather than a personal hobby account. Lure hosts included subdomains of eventrsvp.app such as openai.eventrsvp.app and spacex.eventrsvp.app, with earlier rsvpconnect-events.com infrastructure in the same window. Cloudflare managed challenges sat in front of the forms. Delivery rode Google and Zapier short links, trusted open redirects including an x.com logout redirect, and invitation paths that overlapped GoTo Webinar and Zoom-style event branding Cofense had already flagged in late October 2026.
Selecting Microsoft showed a Kali365 device code on the lure. Victims pasted that code into the real Microsoft device-authorization endpoint at login.microsoftonline.com, signed in with the work account, and approved the grant. Microsoft then issued tokens to the operator’s Kali365 session. The flow still redirected into a VoidProxy-fronted Microsoft sign-in host positioned to capture password material and a session cookie on top of those tokens.
Selecting Google opened a VoidProxy reverse proxy of the real Google sign-in. Push states the proxy relays Google’s real sign-in flow and captures the password, the MFA response, and the session cookie. Push describes Kali365 and VoidProxy as separate phishing-as-a-service platforms built to bypass or defeat MFA checks. No victim organization names, account counts, or confirmed post-compromise actions appear in that reporting. The identity outcome is still clear: operators left the RSVP holding live cloud sessions for employees who thought they were claiming conference perks.
Earlier in 2026, ANY.RUN’s look at real events.zoom.us partner-summit pages for Meta, OpenAI, and Anthropic already showed dual branching into device-code phishing and adversary-in-the-middle flows. Push ties ZoomMock and later MSDRoot kits to the November eventrsvp.app wave through shared domains and tradecraft. The attack-chain narrative and kit lineage sit in the companion piece on legacymfa.sucks: fake AI event RSVPs stole Microsoft and Google tokens.
Where passwords and relayable second factors failed
VoidProxy’s Google path is classic adversary-in-the-middle abuse of a workforce login. An adversary-in-the-middle attack is a reverse proxy between the user and the real identity provider that presents a convincing sign-in while capturing whatever the user types and whatever the IdP returns. Every transferable factor on that path (password plus the second-factor response Push describes) is visible to the proxy. The session cookie minted at the end is replayable from attacker infrastructure without repeating the prompt. The same cookie-capture pattern applied on the VoidProxy Microsoft hop after Kali365.
Kali365’s Microsoft leg is a different failure mode. Device-code phishing coaches the victim to complete a real sign-in on the vendor’s own origin after pasting a code the attacker controls. The IdP mints access and refresh material for the attacker’s waiting client. Correct passwords and completed second-factor approval on Microsoft’s genuine host still bind tokens to Kali365. Push does not name a specific MFA product beyond second-factor relay or approval language. The documented Microsoft failure did not require breaking cryptography on a fake origin. It required a user willing to finish a legitimate device grant for an event that was never real.
One Google Workspace or Microsoft 365 work session can reach mail, files, chat, and integrated SaaS until those sessions are revoked.
| Path | What the user did | What operators received |
|---|---|---|
| Microsoft via Kali365 | Pasted device code on real deviceauth | Tokens bound to attacker Kali365 session |
| Microsoft via VoidProxy hop | Signed in through reverse proxy | Password material and session cookie |
| Google via VoidProxy | Full sign-in through reverse proxy | Password, MFA response, session cookie |
Detection after the fact is revoke-and-hunt work against sessions that were never authorized for an RSVP page. Wrong-password spikes miss both branches. Device-code grants succeed with correct credentials on the real host. AiTM cookies arrive after successful proxied MFA.
What device-bound phishing-proof authentication changes on each path
The two initial-access paths respond differently. On VoidProxy’s reverse-proxy Google and Microsoft sign-ins, the stolen material is still a relayable password and a transferable second-factor response ahead of a finished attacker-controlled session. A phishing-proof IdP that demands a device-bound private key never sends a secret the proxy can copy. The signature is origin-bound to the real enterprise login host. A spoofed or proxied registration page is the wrong origin and has no transferable OTP, SMS code, email code, or push approval to replay. The Prevention, Not Detection model removes that credential-phase surface before a session exists for the kit.
On Kali365’s device-code branch, Microsoft already completed a real interactive sign-in on login.microsoftonline.com before tokens landed in the operator session. Login cryptography that the user satisfied on the genuine host does not un-issue those OAuth tokens. Phishing-proof factors on that real ceremony still produce a successful grant for whoever holds the device code. Stopping that leg is Conditional Access (or equivalent) policy that blocks or tightly scopes the device-code flow for workforce accounts, plus immediate revoke of access and refresh tokens when a lure is confirmed. Shortening token lifetime alone does not prevent the grant. It only forces the operator to re-steal sooner, and an attacker who registers their own authenticator during the window no longer cares about lifetime.
Session cookies captured through VoidProxy after a completed proxied login are post-sign-in artifacts of that failed credential path. When the reverse-proxy login cannot finish because no relayable factor exists, that cookie is not issued to the operator. Cookies or tokens harvested later by malware on a machine that already held a legitimate session are a different, harder problem. Endpoint compromise after a good login is residual containment work, not something any login MFA undoes.
| Control | Against VoidProxy AiTM login | Against Kali365 device-code tokens |
|---|---|---|
| Password + OTP/push | Relayable through proxy | Still completes real device grant |
| Passkeys at login only | Origin-bound login helps | Grant can still finish on real host |
| Phishing-proof device-bound MFA | Blocks transferable relay | Does not un-issue post-grant tokens |
| Device-code OAuth block + revoke | N/A to pure password AiTM | Primary control for this leg |
Passkeys harden the login act on a hostile origin
Passkeys and WebAuthn put a key pair on the device, keep the private key in hardware, and produce an origin-bound signature at authentication. That design is origin-bound at login. On a generic enterprise identity provider that issues only hardware-bound signatures, a reverse proxy cannot steal a reusable secret the way VoidProxy stole passwords and MFA responses. That is real progress against the Google AiTM branch in this campaign.
Key Takeaways for Defenders
- Block or tightly restrict OAuth device-code flow for corporate Microsoft accounts so a pasted RSVP code cannot bind tokens to attacker infrastructure.
- Replace passwords, OTP, SMS, email codes, and approval push on Google Workspace and Microsoft workforce logins with device-bound, origin-bound signatures that a reverse proxy cannot relay.
- Treat VoidProxy-class cookie theft as the product of a completed proxied login; remove the transferable factors so the kit does not receive a session cookie.
- Revoke access and refresh tokens immediately when fake-event infrastructure is confirmed, and treat every session issued from that lure as live until revoke is confirmed.
Fake AI-brand RSVPs will keep recycling trusted short links and real IdP HTML. The durable answer is fewer transferable factors on every workforce path and no silent device-code grants for accounts that only need interactive, hardware-bound sign-in.
FAQ
Would phishing-proof MFA have stopped the entire Kali365 and VoidProxy campaign?
Phishing-proof MFA would have stopped the VoidProxy AiTM branches of the fake OpenAI, SpaceX, and Anthropic event campaign by removing relayable passwords and second-factor responses from Google and secondary Microsoft work logins, so the reverse proxy could not finish those sessions or mint cookies for the operator. It would not un-issue OAuth tokens Microsoft already granted to Kali365 after victims completed real device authorization on login.microsoftonline.com. That Microsoft leg still needs device-code flow controls and explicit token revoke.
How do you prevent Kali365-style device-code theft on Microsoft work accounts?
You prevent Kali365-style device-code theft on Microsoft work accounts by disabling or Conditional Access-blocking the device-code OAuth grant for users who do not need it, monitoring for unexpected device-auth completions, and revoking access and refresh tokens when a fake-event lure is confirmed. Completing MFA on Microsoft’s real deviceauth page still binds tokens to whoever holds the code, so stronger login cryptography alone does not cancel that grant.
How does device-bound authentication stop VoidProxy on Google sign-in?
Device-bound authentication stops VoidProxy on Google workforce sign-in because the private key never leaves the enrolled device and the signature is bound to the real IdP origin. VoidProxy can mirror Google’s HTML, but it cannot obtain a non-relayable hardware signature meant for the legitimate host, and it receives no OTP, SMS code, or push approval to capture. According to Push Security, VoidProxy’s value was capturing password, MFA response, and session cookie through a relayed real sign-in; those transferable steps disappear when only origin-bound device keys remain.
Are passkeys enough against this fake AI event pattern?
Passkeys are origin-bound at login and would materially harden the VoidProxy Google path by replacing typed secrets with origin-bound signatures on a hardware-backed key. They do not cancel Kali365 tokens issued after a completed real Microsoft device-code grant. Device-code OAuth policy is still required for that leg of the dual path Push documented.