Free ChatGPT Pro and Grok seats were the bait. According to Push Security on 5 November 2026, fake registration pages for OpenAI, SpaceX, and Anthropic events steered employees to sign in with work accounts, then used Kali365 device-code phishing for Microsoft tokens and VoidProxy reverse-proxy phishing for Google (and secondary Microsoft) passwords, live MFA responses, and session cookies. No victim company names or confirmed follow-on actions appear in that reporting, but the login mechanics are clear: workforce identity was the prize, not a consumer app account.

How the OpenAI SpaceX Anthropic RSVP lures landed

Push tracked the campaign through late October 2026 after blocking customer-browser hits on the fake event pages. Lure domains included subdomains of eventrsvp.app such as openai.eventrsvp.app and spacex.eventrsvp.app, plus earlier rsvpconnect-events.com infrastructure. Cloudflare managed challenges sat in front of the pages. Delivery rode Google and Zapier short links, trusted open redirects (including an x.com logout redirect), and overlapping GoTo Webinar / Zoom-style invitation paths Cofense had already flagged around 29 October 2026.

The pages did not ask for a personal Gmail hobby login. They required work-account sign-in to “register,” dangling rewards such as free ChatGPT Pro or Grok access. That single design choice put corporate Entra ID and Google Workspace identities on the line. Earlier in the same window, kits Push labels ZoomMock and later MSDRoot served Anthropic-at-AWS-re:Invent and related AI-brand themes. ANY.RUN’s 15 July 2026 look at real events.zoom.us partner-summit pages for Meta, OpenAI, and Anthropic already showed the dual branch into device-code phishing and adversary-in-the-middle flows. Push ties that lineage to the November write-up through shared domains and tradecraft.

An adversary-in-the-middle attack is a reverse proxy that sits between the user and the real identity provider, presenting a convincing login while capturing whatever the user types and whatever the IdP returns. Device-code phishing is different: the victim completes a real sign-in on the vendor’s own page after pasting a code the attacker controls, so the IdP mints tokens for the attacker’s waiting client.

What Kali365 did with Microsoft device codes

Choosing Microsoft on the lure did not open a lookalike password box first. Kali365 displayed a device code on the fake event page. Victims were coached to open the real Microsoft device-authorization endpoint at login.microsoftonline.com/common/oauth2/deviceauth, paste that code, sign in with the corporate account, and approve the grant. According to Push, Microsoft then issued tokens to the operator’s Kali365 session. The authentication ceremony happened on Microsoft’s genuine origin. The binding of the resulting access and refresh material went to infrastructure the victim never owned.

After that grant, the lure path still redirected into a VoidProxy-fronted Microsoft sign-in host (Push cites patterns such as continue-to-office-.ggleventapi.icu). That second hop was positioned to capture password material and a session cookie on top of the tokens Kali365 already held. Kali365 and VoidProxy are separate phishing-as-a-service platforms. Push describes both as built to bypass or defeat MFA checks. The Microsoft device-code leg did not need to break the cryptography of a second factor on a fake origin. It needed a user willing to complete a legitimate device grant for an event that was never real.

Push does not name a specific MFA product on those work accounts beyond second-factor relay or approval language. It also does not publish account counts, named employers, or what operators did with the sessions after collection. The documented failure is still severe: one pasted code on the real Microsoft device page was enough for tokens to land in Kali365.

Path User action What operators received
Microsoft via Kali365 Paste device code on real Microsoft deviceauth Tokens bound to attacker Kali365 session
Microsoft via VoidProxy follow-on Sign-in through reverse proxy Password material and session cookie
Google via VoidProxy Full sign-in through reverse proxy Password, MFA response, session cookie

How VoidProxy finished Google and cookie theft

Selecting Google opened a VoidProxy reverse proxy of the real Google sign-in flow. Push states the proxy relays Google’s real sign-in and captures the password, the MFA response, and the session cookie. That is classic credential-phase AiTM against a workforce Google login: every transferable factor the user presents is visible to the proxy, and the 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. Session cookies taken through an attacker-controlled reverse proxy are not a separate mystery after a successful proxied login. They are the natural product of completing authentication on a path the operator owns. One Google Workspace or Microsoft 365 work session can reach mail, files, and integrated SaaS until those sessions are revoked.

Push’s November analysis does not document ransomware, bulk exfil volumes, or mailbox-export frameworks for this specific wave. The identity outcome still matches other industrial PhaaS runs: operators leave the lure holding live cloud sessions for corporate users who thought they were claiming conference perks. Short links and trusted redirects reduced the chance a busy employee paused on the domain. The Cloudflare gate added a veneer of legitimacy before the registration form ever asked for the work account.

Compared with helpdesk password-reset social engineering or pure push-fatigue bombing, this campaign’s Microsoft leg outsourced the hard part to Microsoft’s own device-code UX. The Google leg outsourced credibility to Google’s real HTML, CSS, and MFA challenge, only relocated behind VoidProxy. Either way, the user performed a normal-looking work login under event branding that was false from the first pixel.

Defenders who only hunt “wrong-password spikes” miss both branches. Device-code grants succeed with correct passwords and completed MFA on the real host. AiTM cookies arrive after successful proxied MFA. Detection after the fact is revoke-and-hunt work against sessions that should never have been minted for an RSVP page. Prevention has to remove transferable factors and unsafe OAuth device grants from workforce login paths before the lure is clicked.

The defensive side of this case is covered in the related article on mfa2point0.com: how phishing-proof MFA stops AiTM relay and why device-code grants still need a block.

FAQ

How did the fake OpenAI and SpaceX event pages steal work accounts?

The fake OpenAI, SpaceX, and Anthropic event pages stole work accounts by forcing corporate Microsoft or Google sign-in on the RSVP form. According to Push Security, Microsoft selections used Kali365 device codes pasted into real login.microsoftonline.com device authorization so tokens issued to the attacker session, while Google selections used a VoidProxy reverse proxy that captured passwords, MFA responses, and session cookies.

What is Kali365 in this campaign?

Kali365 is the phishing-as-a-service platform Push Security says handled the Microsoft branch of the fake AI-event campaign. It showed a device code on the lure, waited for the victim to complete device authorization on Microsoft’s real login host, and received the tokens Microsoft issued to the operator’s Kali365 session.

What is VoidProxy doing on the Google path?

VoidProxy reverse-proxied the real Google sign-in for victims who chose Google on the fake event pages. Push reports it captured the password, the MFA response, and the session cookie during that relayed flow, and it also fronted a secondary Microsoft sign-in hop after some Kali365 completions.

Did Push name the victim companies or how many accounts were hit?

Push Security’s 5 November 2026 analysis does not name victim organizations, publish account or record counts, or confirm post-compromise actions for the Kali365/VoidProxy fake-event wave. The public record covers lure infrastructure, kit names (ZoomMock/MSDRoot), and the Microsoft versus Google login mechanics.

Is this the same pattern as earlier Zoom and GoTo event lures?

Push links the November OpenAI/SpaceX/Anthropic eventrsvp.app wave to earlier ZoomMock activity on rsvpconnect-events.com and to July 2026 events.zoom.us partner-summit lures that already branched into device-code and AiTM phishing. Cofense’s late-October GoTo Webinar Anthropic-themed invitations sit in the same reporting window as adjacent invitation infrastructure, not as a separately proven operator identity.