Fake OpenAI and SpaceX event pages phish Microsoft, Google logins
Push Security details a phishing campaign using fake OpenAI, SpaceX and Anthropic event registrations to run Microsoft device-code and Google AiTM attacks against corporate accounts.
A phishing campaign is impersonating event invitations from OpenAI, SpaceX and Anthropic to steal corporate Microsoft and Google logins, Push Security reported. The lures route victims into two different token-theft techniques — Microsoft device-code phishing and a Google adversary-in-the-middle (AiTM) proxy — so stolen passwords and session cookies, not just credentials, land in the attacker's hands. If your users sign in to third-party "events" with work accounts, this is a session-hijack risk, not a password-reset one.
The lure
The fake registration pages dangle AI-branded perks:
- An OpenAI page — "ChatGPT for Social & Digital Marketing Teams" — offering three months of ChatGPT Pro.
- A SpaceX page for a private event on November 12, offering three months of Grok Heavy or a "SpaceX share."
- An Anthropic page tied to AWS re:Invent 2026, already taken down by the time of reporting.
Each page tells the visitor to sign in with a work Microsoft or Google account to register and claim the reward. Push Security says it detected and blocked visits to the pages in late October 2026.
How the theft works
The campaign leans on two phishing-as-a-service platforms, Kali365 and VoidProxy, and Push Security assesses that a single operator runs two custom kits it tracks as ZoomMock and MSDRoot.
- Microsoft accounts (Kali365): device-code phishing. The page generates a legitimate Microsoft device-authentication code and tells the victim to enter it on Microsoft's real login endpoint. That abuses the OAuth 2.0 device-authorization grant to mint access and refresh tokens without ever handling the password. The flow can then redirect to a VoidProxy Microsoft login page that also captures the password and session cookie.
- Google accounts (VoidProxy): an AiTM reverse proxy served in a pop-up. It relays the real Google login in real time, capturing the credentials, the MFA response and the resulting session cookie — enough to replay an authenticated session and bypass MFA.
Because device-code and AiTM flows both yield live session tokens, MFA alone does not stop them.
Evasion and infrastructure
A Cloudflare CAPTCHA ("Confirm you're not a robot") sits in front of the pages to screen out automated scanners before the phishing content loads. Victims are funneled through trusted redirectors, including Google and Zapier URL shorteners. Push Security says the initial delivery vector is unclear; earlier iterations of this lure family abused GoTo Webinar mail systems and legitimate Zoom event pages. Push Security's writeup carries the full technical breakdown and indicators; the public reporting does not list domains or hashes, so pull them from upstream rather than guessing.
Attribution
No named actor. Push Security attributes the two custom kits to one operator but does not tie the campaign to a tracked group, and the delivery infrastructure is shared PhaaS tooling rather than a fingerprintable crew. Related lure families have been reported before it — ANY.RUN flagged abused Zoom event pages impersonating Meta, OpenAI and Anthropic in July 2026, and Cofense documented fake Anthropic invitations sent through GoTo Webinar infrastructure.
Action checklist
- Alert users now: a third-party "event" that asks for a work Microsoft or Google login to register is the tell. Perks tied to OpenAI, SpaceX or Anthropic are the current bait.
- In Entra ID, restrict the OAuth 2.0 device-authorization flow — scope it to the few platforms that need it with a Conditional Access policy, or block it where it is unused.
- Hunt for anomalous device-code sign-ins and token grants (
OAuth2:DeviceCode) in sign-in logs, especially from unmanaged devices or unexpected geographies. - For Google Workspace, require phishing-resistant MFA (passkeys / FIDO2) for admins and high-value users; AiTM proxies defeat OTP and push MFA but not hardware-bound credentials.
- Revoke and reissue sessions for any user who entered credentials or a device code on one of these pages; a cookie replay survives a password change.
Context
This is the same playbook we have covered twice this quarter: AI-branded event invites as a credential-phishing wrapper. Cisco Talos tied UAT-11985 to Google AiTM lures with event-invite bait, and Proofpoint documented TA419 phishing US AI-policy circles. The AI-event theme keeps working because the perks are plausible and the sign-in prompt looks routine — and the move to device-code and AiTM token theft means the industry's MFA advice no longer covers the gap on its own.