The short version: Push Security reported fake AI-themed event pages targeting work accounts. One Microsoft path uses a real device sign-in page and then a separate deceptive sign-in page. If you approved a sign-in you did not initiate, contact your IT team promptly through a trusted channel.
The AI Event Invite That Could Hand Hackers Your Work Account
Researchers found fake AI event pages designed to lure employees into signing in with Microsoft or Google work accounts. In one version, the Microsoft sign-in page itself is genuine, but the request starts with an attacker.

Jump to a section
The invitation promises something people at work might actually want: an AI briefing, a new tool or access to a premium service. The registration asks you to sign in. A familiar Microsoft page appears.
That real sign-in page is the uncomfortable part. A genuine address can tell you where you are signing in. It cannot, by itself, tell you who started the request or where the resulting access will go.
On November 5, Push Security reported that it had detected and blocked visits to fake event registration pages on customers' browsers in late October. The pages impersonated events associated with OpenAI, SpaceX and Anthropic and offered rewards in exchange for work-account sign-ins. Push described malicious pages and attempted credential capture; it did not disclose a count of compromised accounts. The impersonated companies did not run the fraudulent events described by the researchers. We checked the reporting and provider guidance on November 16.
First, a useful promise
One fake page advertised an OpenAI-themed livestream and access to a paid AI subscription. Another promoted a SpaceX briefing with a choice of rewards. Push also found an Anthropic-themed page, which it said was no longer live when researchers investigated.
The subject is not an accident. An invitation about a technology people are being encouraged to learn can look like a professional opportunity. Asking for a work account may even seem consistent with a work event.
That is the first trust transfer: interest in a real company becomes confidence in a page claiming to represent it. A familiar brand or a link delivered through a legitimate service does not establish who created the event. Push traced links through familiar link-shortening services and said earlier invitations used legitimate webinar services. BleepingComputer reported that the initial distribution of the links it discussed was unclear.
The real Microsoft page and the wrong device
One of the Microsoft techniques described by Push uses a legitimate feature called device-code sign-in. It was designed for situations where typing a password on the device you want to use is difficult, such as a shared conference-room device. A person enters a code on a separate device to authorize the original one.
In the phishing version, the attacker starts the sign-in on a device or application they control. The fake event page supplies the code and directs the person to Microsoft's real sign-in page. If the person enters the code and approves the request, the attacker can receive access associated with that sign-in.
In the sequence Push examined, the page then directed the visitor to a separate fake Microsoft sign-in page designed to capture a password and session information. This second step was not the genuine Microsoft page. The attacker could attempt to obtain access through both stages; the report does not establish that every target completed them.
Microsoft explains that the real sign-in page can make this attack harder to recognize. The detail to question is not simply whether the address belongs to Microsoft. It is whether you initiated the device sign-in and recognize the device or application receiving access.
Imagine someone hands you a real key and asks you to open a real door. Those facts matter. But you still need to know who is waiting on the other side. This is an analogy, not an account of a specific victim.
A deceptive sign-in page can target both providers
Push says the Google option used a separate service to create a proxy, a system positioned between the person and the real sign-in service. It can relay authentication and capture credentials or session information as they pass through. Push says that same proxy service also powered the later fake Microsoft page, after the device-code step.
A session is the state that lets a service recognize you as signed in. If an attacker obtains usable session information, the problem may extend beyond the password itself. Push's report describes the techniques used by the pages it investigated; it does not mean every fake event uses both methods or that every attempt succeeds.
Two-factor authentication remains valuable. Some methods, including phishing-resistant security keys and passkeys when correctly used, provide stronger protection against certain phishing techniques. But a person can still be deceived into authorizing a device-code request they did not initiate. No single safeguard should be treated as a complete account-security plan.
The warning behind the warning
The Microsoft path in Push's report used a phishing service called Kali365. Months earlier, on May 21, the FBI warned about Kali365, describing a service that can obtain Microsoft 365 access tokens and help attackers enter accounts without directly capturing a password.
The FBI warning is background on that service, not evidence that the bureau investigated this specific event campaign. Together, the sources show how a reusable phishing service can be paired with a timely lure. They do not establish how many people lost access or whether the same group created every component.
There is a parallel with FaceStealer's fake meetings: both exploit the familiarity of an online work routine. The operations and mechanisms are different. In FaceStealer, the FBI described fake meetings used to persuade targets to install malicious software. Here, the AI event pages described by Push target authentication and account sessions.
What to do if you received an invitation
- Received a message but did not open the link: Do not use the invitation to verify itself. Find the event through the company's official site or contact the supposed organizer using a channel you already trust. Report suspicious messages according to your workplace policy.
- Opened the page but did not sign in or approve anything: Leave it and report the link to your IT or security team if it concerns a work account. Opening a page alone does not establish that your account was taken over, although your team may advise further checks.
- Entered a password: Contact your IT team promptly through a trusted channel. From a device you trust, use the provider's official site to change the password as your organization directs. If it is a personal account, change the password there promptly. Change it anywhere else it was reused and tell your IT team if a work password was affected.
- Entered a device code or approved a sign-in you did not initiate: Treat it as urgent. Contact your IT or security team immediately and explain what you approved. Preserve the invitation and relevant details for their investigation. A password change alone may not end every existing session, so ask the team to review and revoke sessions and other access where appropriate. Do not send anyone the code or password when reporting the incident.
For a personal Google account, Google's unfamiliar activity guidance explains how to review security events and devices. If you cannot access a personal Microsoft account, our Microsoft account recovery guide explains where to start. Work accounts may have different recovery controls, so involve the account administrator.
What organizations can change
Training can teach people to pause before authorizing an unfamiliar request. It should also make reporting easy and blame-free: a rapid report gives defenders a better chance to review access while the details are fresh.
Microsoft recommends restricting device-code flow where possible. Before applying a broad block, administrators should identify legitimate dependencies, test the effect and scope any needed exceptions. Some Teams devices and other tools rely on that flow.
Organizations can also review suspicious sign-ins and sessions, reduce unnecessary access, and use phishing-resistant authentication where supported. In a suspected Microsoft 365 compromise, Microsoft recommends steps including revoking active sessions and reviewing the account. Its emergency access guidance notes that some application sessions and access tokens need separate handling or may not be invalidated immediately. These controls are layers, not a guarantee.
Our analysis of stolen AI keys and software access explores a different version of the same broader question: which identities have access, and how can that access be stopped? The incidents are separate, but both show why changing one person's password is not always the whole response.
Who is the sign-in really for?
A convincing invitation can borrow trust from a recognizable company, a familiar meeting routine and a real sign-in page. Each detail may be genuine or plausible while the request as a whole is deceptive.
The useful pause comes before authorization: did you start this request, do you recognize what is receiving access, and have you confirmed the event independently? A trustworthy page is one part of the answer. It is not the whole answer.