Anyone with a work Microsoft or Google account who receives an invite to an OpenAI, SpaceX or Anthropic event, especially one promising free ChatGPT Pro or Grok Heavy access, should treat it as phishing.
Researchers at Push Security found fake pages for events that do not exist: an OpenAI Academy livestream, a private SpaceX briefing on AI marketing, and an Anthropic presentation at AWS re:Invent. Visitors were offered a free ChatGPT Pro subscription, Grok Heavy access or even SpaceX shares, but had to sign in with Microsoft or Google first.
The two sign-in paths work differently. Choose Microsoft, and a service called Kali365 starts device-code phishing. The victim receives a code, enters it on Microsoft's real device login page and approves the request, and Microsoft then issues an authentication token to the attacker's session. The victim is next sent to a fake Microsoft sign-in page run by VoidProxy, which captures more credentials and session cookies.
Choose Google, and VoidProxy serves a reverse-proxied copy of Google's sign-in page. It relays the login as it happens while capturing the password, the MFA response and the session cookie. That hands the attackers an authenticated session without having to defeat MFA. Both tools are phishing-as-a-service platforms, meaning criminals rent them instead of building their own.
Push does not know how the links are distributed. It did find the attackers using Google and Zapier to shorten links and redirect visitors, which helps them slip past security scanners and URL reputation systems.
The device-code step is the most awkward part for defenders. The victim types a code into a genuine Microsoft page, so there is no suspicious domain to catch at that moment, and the usual advice to check the address bar does little. Push said the campaign is built to get around the controls most organizations rely on against phishing.
The lure is also familiar. ANY.RUN reported fake invitations impersonating Meta, OpenAI and Anthropic on legitimate Zoom event pages in July 2026, and Cofense later saw fake Anthropic invites sent through GoTo Webinar. Each borrows a trusted brand and a trusted platform. The harder question is how much a sign-in flow should trust a code that a user types in on request.