Attackers use fake AI event invitations to hijack Microsoft and Google sign-ins

Signing in to claim a free AI subscription can hand an attacker access to a corporate account—even when part of the process happens on Microsoft’s real website. Push Security disclosed a campaign on November 5 that uses fake OpenAI, SpaceX and Anthropic event invitations to draw employees into Microsoft and Google sign-in traps.

Push said it detected and blocked visits to the fraudulent registration pages in customer browsers in late October. The campaign’s operator built the event pages, then connected them to two rented phishing services, Kali365 and VoidProxy. Both services are designed to bypass or defeat multifactor authentication, according to the researchers.

A work login becomes the price of admission

One page copied the OpenAI Academy events site and advertised a November 4 livestream called “ChatGPT for Social & Digital Marketing Teams.” Its promised reward was three months of ChatGPT Pro. A SpaceX version advertised an invite-only November 12 “Starship Product Briefing” about AI marketing and Grok, offering three months of Grok Heavy or a purported SpaceX share.

These were fraudulent invitations, not company offers. Both pages required a work account and presented “Continue with Microsoft” and “Continue with Google” buttons. They used the same code, translated themselves into the visitor’s language and sent each step to the operator’s collection server. An Anthropic-branded version advertised an appearance at AWS re:Invent; that page was no longer live when Push investigated.

Two buttons, different routes to account access

The Microsoft route begins with a device code: a code the visitor is told to copy into a login page. Kali365 supplies that code, and the fake event page opens Microsoft’s genuine device-login website in a pop-up. When the victim enters the code, signs in and approves the request, Microsoft issues authentication tokens to the operator’s Kali365 session.

The trap does not end there. The lure then claims the visitor is signed in and opening Microsoft 365. Instead, it redirects to a VoidProxy Microsoft sign-in page that captures the password and session cookie—the browser data used to maintain a signed-in session. That gives the operator a second capture after the device-code approval.

Google users encounter a different mechanism. Their button opens a pop-up backed by VoidProxy, which acts as an intermediary between the victim and Google’s real sign-in process. It relays authentication while collecting the password, multifactor authentication response and session cookie. Rather than simply presenting a static imitation login form, this route captures data as the real authentication proceeds.

Trusted links help conceal the destination

Push’s attack-chain analysis describes invitation links passing through Google short links or Zapier’s zpr.io service. One examined chain also used an X logout redirect. Earlier invitations came through GoTo Webinar’s mail sender or real Zoom event pages. Those legitimate services were parts of the delivery chain, not evidence that the destination event was genuine.

A Cloudflare challenge stopped most public scans before they reached the lure. The VoidProxy Microsoft proxies showed scanners real business-news sites, while the operator rotated domains every few weeks. Push cautioned that its published domain indicators therefore have limited value alone; they are more useful for searching historical logs for earlier visits.

The researchers linked the newer event-page kit, which they call MSDRoot, to an earlier fake-Zoom kit named ZoomMock with medium-high confidence, mainly because both used one primary domain. The kits share no code. Push found no link between the developers of Kali365 and VoidProxy and assesses that the campaign operator is most likely a customer of both services.