Device Code Phishing Uses Fake OpenAI, SpaceX Events to Hijack Logins
Fake OpenAI, SpaceX and Anthropic event pages send Microsoft users into Kali365 device code phishing and Google users into a VoidProxy AiTM proxy, stealing tokens and session cookies that bypass MFA.
Threat card
- Threat type
- Phishing & Scams
- Disclosed
- Nov 5, 2026
- Affected
- Corporate Microsoft Entra ID (Microsoft 365 and connected SSO apps) and Google accounts; victim count not disclosed
The short version
- Fake OpenAI, SpaceX and Anthropic event pages ask visitors to sign in with work accounts to claim AI subscriptions or a SpaceX share.
- Microsoft users are pushed into Kali365 device code phishing, then a VoidProxy login page that also grabs the password and cookie.
- Google users get a VoidProxy adversary-in-the-middle pop-up that steals credentials, MFA responses and session cookies.
- Push Security blocked visits in late October 2026 and links two custom kits, ZoomMock and MSDRoot, to one actor.
A phishing operation is using fake invitations to OpenAI, SpaceX and Anthropic events to steal corporate Microsoft and Google accounts, and it pairs two phishing-as-a-service platforms to do it. Push Security says it detected and blocked visits to the fake registration pages in late October 2026. Microsoft users are hit with device code phishing through Kali365, while Google users land in a VoidProxy adversary-in-the-middle (AiTM) proxy. Both routes defeat ordinary MFA. Defenders should restrict the device code flow and hunt for token and cookie replay now.
What happened
Attackers built fake event registration pages that offer free premium AI subscriptions or a “SpaceX share” to people who sign in with a work account. Choosing Microsoft triggers a device code request from Kali365; choosing Google opens a VoidProxy reverse proxy. Either way the attacker walks away with tokens or session cookies that bypass MFA.
The campaign was reported by BleepingComputer, citing a new Push Security report. Push tracks two custom phishing kits, ZoomMock and MSDRoot, that render the fake event pages and hand victims to the Kali365 and VoidProxy services. The researchers believe one actor runs both kits because they share infrastructure and targeting.
The lures are specific. One page advertises an OpenAI event called “ChatGPT for Social & Digital Marketing Teams” with three months of ChatGPT Pro for attendees. Another promotes a private SpaceX event on 12 November, offering three months of Grok Heavy or a “SpaceX share”. A third page, impersonating an Anthropic event at AWS re:Invent 2026, had already been taken down.
Timeline
| Date | Event |
|---|---|
| 11 September 2025 | Okta Threat Intelligence publishes its analysis of the VoidProxy phishing-as-a-service platform. |
| July 2026 | ANY.RUN documents a similar campaign abusing legitimate Zoom event pages to promote fake Meta, OpenAI and Anthropic events leading to device code and AiTM phishing. |
| Recent weeks | Cofense reports Anthropic event invitations sent through GoTo Webinar’s email infrastructure and redirected to fake registration pages. |
| Late October 2026 | Push Security detects and blocks visits to the fake OpenAI, SpaceX and Anthropic event pages. |
| 5 November 2026 | The campaign is reported publicly. |
| 12 November 2026 | Date of the fake “private SpaceX event” used as a lure. |
How the attack works
This is a two-platform kill chain built for identity theft, not malware delivery. Nothing is installed on the victim’s device; the prize is an authenticated session.
Delivery and redirection
How the first links are distributed is not yet clear. Push found that victims reach the fake registration page through links on trusted services, including Google and Zapier URL shorteners. Earlier waves were sent through GoTo Webinar’s email systems or legitimate Zoom event pages. Using trusted redirectors means the first URL a mail filter sees belongs to a reputable domain.
Scanner evasion
Before the event page loads, the site shows a Cloudflare CAPTCHA asking visitors to confirm they are not a robot. This stops automated security crawlers from reaching and classifying the phishing content. Okta’s earlier research on VoidProxy described the same pattern: Cloudflare CAPTCHA checks, a Cloudflare Worker that likely filters traffic, and a harmless “Welcome!” page served to scanners.
Microsoft path: Kali365 device code phishing
When the victim clicks “Continue with Microsoft”, the page asks Kali365 to generate a Microsoft device authentication code and shows it in a form. The victim is told to enter the code on Microsoft’s real device login page and approve the request. That approval completes an OAuth 2.0 device authorization grant for the attacker’s session, so Kali365 receives tokens for the victim’s Microsoft Entra account without ever seeing a password. Those tokens can reach Microsoft 365 and connected single sign-on apps such as Salesforce, Google Workspace, Dropbox, Slack, Zendesk and Atlassian.
After the code is approved, the victim is redirected to a VoidProxy Microsoft login page, which captures the password and session cookie as a second layer of theft.
Google path: VoidProxy AiTM
Choosing “Continue with Google” opens a pop-up with a Google login page served by VoidProxy. The platform sits between the browser and Google, relays the real login forms and records credentials, MFA responses and the session cookie Google issues. Okta’s analysis shows the panel lets VoidProxy customers watch victims in real time and receive stolen data through Telegram bot tokens or webhooks, and that accounts federated to Okta are routed to second-stage pages that mimic Okta sign-in.
Impact
With a refresh-capable token or a live session cookie, the attacker could read mail, files and SaaS data as the user, register new sign-in methods or plant mailbox rules. Device code phishing has previously been used for data theft and espionage, according to the reporting.
MITRE ATT&CK mapping
| Tactic | Technique (ID) | How it was used here |
|---|---|---|
| Initial Access | Phishing: Spearphishing Link (T1566.002) | Event invitation links routed through Google and Zapier shorteners, GoTo Webinar and Zoom event pages. |
| Defense Evasion | Impersonation (T1656) | Fake OpenAI, SpaceX and Anthropic event pages promising rewards. |
| Credential Access | Steal Application Access Token (T1528) | Kali365 obtains Entra ID tokens through the device code flow. |
| Credential Access | Adversary-in-the-Middle (T1557) | VoidProxy reverse proxies Google and Microsoft sign-ins. |
| Credential Access | Steal Web Session Cookie (T1539) | VoidProxy captures authenticated session cookies. |
| Credential Access | Input Capture: Web Portal Capture (T1056.003) | Passwords and MFA responses captured on proxied login pages. |
| Defense Evasion, Lateral Movement | Use Alternate Authentication Material: Application Access Token (T1550.001) | Stolen tokens reused against Microsoft 365 and SSO apps. |
| Defense Evasion, Lateral Movement | Use Alternate Authentication Material: Web Session Cookie (T1550.004) | Stolen Google and Microsoft session cookies replayed to skip MFA. |
Who is behind it
The operator is not yet attributed. Push Security assesses that a single threat actor runs both the ZoomMock and MSDRoot kits, based on shared infrastructure and targeting, but no name, alias or nationality has been published. Kali365 and VoidProxy are rented platforms, so the operators of the services and the customer running this campaign may be different people. Okta noted in 2025 that VoidProxy provisions isolated infrastructure per customer, which makes linking activity to one operator difficult.
Affected products and versions
| Platform | Attack used | Data at risk | Population |
|---|---|---|---|
| Microsoft Entra ID (Microsoft 365 and connected SSO apps) | Kali365 device code phishing, then VoidProxy login capture | OAuth access and refresh tokens, passwords, session cookies | Not disclosed |
| Google accounts (Google Workspace) | VoidProxy AiTM pop-up | Passwords, MFA responses, session cookies | Not disclosed |
| Accounts federated to Okta | VoidProxy second-stage Okta lookalike (per Okta’s 2025 research) | Credentials and sessions | Not disclosed |
No software vulnerability or CVE is exploited. The campaign abuses legitimate authentication flows.
Indicators of compromise
Push Security’s domains for this wave were not listed in the public reporting. Until they are, hunt on the behavioural and infrastructure patterns below, which come from the reporting and from Okta’s VoidProxy research.
| Type | Indicator (defanged) | Context |
|---|---|---|
| Lure text | “ChatGPT for Social & Digital Marketing Teams”, “SpaceX share”, Grok Heavy offer | Fake event registration pages |
| Kit names | ZoomMock, MSDRoot | Custom kits feeding Kali365 and VoidProxy |
| Redirectors | Google and Zapier URL shorteners | First hop of the lure link |
| Hostname pattern | login.<phishing_domain>.<tld>, accounts.<phishing_domain>.<tld> |
VoidProxy first-stage pages (Okta, 2025) |
| TLDs | .icu, .sbs, .xyz |
Low-cost TLDs used by VoidProxy front ends (Okta, 2025) |
| Infrastructure | *.workers[.]dev, sslip[.]io, nip[.]io |
Cloudflare Workers and wildcard DNS used by VoidProxy (Okta, 2025) |
Detection and hunting
For the Microsoft path, review Entra ID sign-in logs for the device code authentication protocol. Most organisations have few legitimate device code sign-ins, usually from meeting room devices or CLI tools, so any device code sign-in for a normal user, especially one where the device location differs from the user’s usual location, deserves triage. Follow up with audit logs for new MFA methods, new inbox rules and OAuth consent grants on the same account.
For the Google and Microsoft AiTM path, look for a successful sign-in followed quickly by session activity from a different IP address, ASN or user agent, which is the classic sign of cookie replay. Google Workspace login audit events and Microsoft’s risky sign-in detections both help here.
At the web proxy, flag visits that pass through Google or Zapier short links and land on newly registered domains showing a Cloudflare CAPTCHA, and any user who visits microsoft.com/devicelogin shortly after browsing an unfamiliar site. Email gateways should flag event invitations that require a corporate login to claim a reward.
Mitigation and response
- Block or restrict the device code flow with a Conditional Access authentication flows policy, allowing it only for named accounts or devices that need it.
- Move high-value users to phishing-resistant MFA such as FIDO2 security keys or passkeys, which AiTM proxies cannot relay.
- For any user who interacted with a fake event page, revoke all sessions and refresh tokens, reset the password and review registered MFA methods, inbox rules and OAuth grants.
- Hunt for device code sign-ins and cookie replay patterns across the last 30 days, not just since late October, since similar fake event waves were reported as early as July.
- Require compliant or managed devices for access to sensitive apps so a stolen token or cookie alone is not enough.
- Brief staff that no real event needs a device code or a work sign-in to claim free AI subscriptions or shares.
Analyst take
The interesting detail is not the lure but the plumbing. One page routes Microsoft users to device code phishing and Google users to an AiTM proxy, then stacks a second capture on the Microsoft path for good measure. That is phishing-as-a-service maturing into an identity attack platform, where the operator picks the flow that best defeats each provider’s MFA. AI brands are simply the bait of the season, the same social engineering instinct we saw in the fake IT support intrusions on Microsoft Teams and the recruiter lure behind the Binance root cause report.
Marketing, sales and executive teams who attend vendor events are the obvious targets, and any tenant that still allows device code sign-ins for every user is exposed. If you cannot roll out phishing-resistant MFA quickly, closing the device code flow is the single cheapest control that breaks half of this campaign today. The VoidProxy background here comes from Okta’s analysis of the VoidProxy platform.
What to do now
0 of 6 done
Questions readers ask
What is device code phishing?
It abuses the legitimate OAuth 2.0 device authorization flow: the victim enters an attacker-generated code on Microsoft's real login page, which hands the attacker tokens for the account without exposing the password.
How does VoidProxy bypass MFA?
VoidProxy is an adversary-in-the-middle proxy that relays the real login page and copies the session cookie issued after MFA, so the attacker can reuse the logged-in session.
How do I stop device code phishing in Microsoft 365?
Block or limit the device code flow with Conditional Access, use phishing-resistant MFA, and revoke sessions for any user who entered a code from an untrusted site.
Sources
- BleepingComputer: Fake OpenAI, SpaceX events used to steal Microsoft, Google accountsbleepingcomputer.com
- Okta Threat Intelligence: Uncloaking VoidProxysec.okta.com
One short email when something serious breaks.
Threatspedia covers major cyber threats only, verified against primary sources. Read our editorial policy, or report a correction.


