Security researchers have uncovered a phishing operation that leans on fake AI event invites to trick employees into surrendering corporate Microsoft and Google credentials. The campaign, documented by Push Security, combines device-code phishing with reverse-proxy techniques and multi-factor authentication bypass, according to the firm’s analysis.
Invented Events, Real Credential Loss
The scheme revolves around fabricated invitations to gatherings that do not exist. Among the lures are an OpenAI Academy livestream, a private SpaceX briefing on AI marketing, and an Anthropic presentation at AWS re:Invent. Visitors to the fraudulent pages are promised rewards such as a free ChatGPT Pro subscription, access to Grok Heavy, or even shares of SpaceX. To claim any of these, they are asked to sign in with a Microsoft or Google account.
That sign-in step is the trap. No reward follows; instead, the credentials go to the attackers. Push Security noted that employees’ appetite for the newest AI tools makes them highly effective bait.
Two Platforms, Two Paths to Compromise
The campaign relies on two phishing-as-a-service platforms, Kali365 and VoidProxy. The path a victim takes depends on which account type they choose.
The Microsoft Route
Selecting Microsoft credentials triggers an attack through Kali365. The victim is given a device code and told to enter it on Microsoft’s genuine device login page. After signing in and approving the request, Microsoft issues an authentication token to the attacker’s Kali365 session. The fake event page then sends the victim to a counterfeit Microsoft sign-in page run by VoidProxy, which harvests further credentials and session cookies.
The Google Route
Choosing Google leads to a different method. The victim sees a reverse-proxied copy of Google’s legitimate sign-in page, operated by VoidProxy. The proxy relays the authentication process while capturing the password, MFA response, and session cookie. This lets the attackers obtain an authenticated session without having to defeat MFA directly.
Distribution and Evasion
Push Security does not know exactly how the links are spread. Phishing emails, social channels, or Telegram groups are plausible vectors. What is clear is that the attackers use Google and Zapier to shorten links and redirect visitors, a tactic meant to slip past security scanners and URL reputation systems.
The firm said the combination of evasion techniques and MFA-bypassing payloads is built to get around or through the controls most organizations depend on to stop phishing today.
A Growing Pattern
AI-themed phishing has become a recurring problem. Previous campaigns have involved Claude installation pages promoted through search ads and malicious links disguised as legitimate AI content, including shared conversations, downloadable artifacts, and custom GPTs. In July 2026, ANY.RUN reported fake event invitations impersonating Meta, OpenAI, and Anthropic that were hosted on legitimate Zoom event pages. More recently, Cofense observed a campaign built around fake Anthropic invitations distributed through GoTo Webinar.
As AI tools become more deeply embedded in personal and business life, threat actors are expected to keep exploiting them. For employees, the practical warning is simple: an unsolicited invitation to an exclusive AI event, especially one that asks for a work login to unlock perks, deserves scrutiny rather than a click.
