Field Notes / Phishing
The fake Anthropic invite passes the hover test. You sign in on Microsoft’s real page.
Invitations to “private” Anthropic, OpenAI and SpaceX events are landing in work inboxes, with free AI subscriptions as the bait. The usual advice says check the link. In the Microsoft version of this attack the link is fine, the sign-in page is genuine, and the attacker still walks away with your account. Here is how it works and what actually stops it.
Yes, the convincing “Anthropic event” emails going round are phishing, and so are the OpenAI and SpaceX versions that come from the same operator.
Each one invites you to a private briefing, promises a free subscription for turning up, and asks you to confirm with your work account, which is the whole point: what you are really confirming is a sign-in that hands the attacker your Microsoft 365 or Google Workspace account. Your MFA does not stop it.
Cofense has tracked the Anthropic-branded wave since September 2026 Cofense , and on 5 November Push Security published the full chain, along with OpenAI and SpaceX lures from the same kits aimed squarely at marketing staff Push Security . This week a newer version was shared on LinkedIn: a “Claude Fable 5.2” event for European marketing teams on 19 November, with six months of Claude Pro and $500 in tokens for everyone who confirms with a work email.
The advice under that post was the advice everyone gives (hover over the button, read where it really goes).
Against half of this campaign, it does nothing.
01 Four screens from invite to stolen account
Push walked through the Microsoft route step by step, and the striking thing is how ordinary it looks: four screens, none of them broken, none of them asking for anything a busy person would hesitate over Push Security .
- The invitation. It arrives by email. Earlier waves came from GoTo Webinar’s own sending address, customercare@gotowebinar.com, which most mail filters trust Cofense . The link is a Google or Zapier short link.
- “Confirm you’re not a robot.” A real Cloudflare check, there mainly to keep security scanners from seeing what comes next.
- The event page. A tidy registration page with two buttons: “Continue with Microsoft” and “Continue with Google”.
- “Confirm access.” Choose Microsoft and the page shows a short code and an “Open Microsoft” button, which opens Microsoft’s genuine device sign-in page on login.microsoftonline.com. You paste the code, sign in, approve the MFA prompt on your phone, and see “Signed in”.
Step four is the attack.
A device code is how you sign in a smart TV or a conference-room screen: the device shows a code, you approve it on your phone, and the device gets your session. Here the “device” is the attacker’s server, and the session it receives is as good as yours. Microsoft calls this device code phishing and has been warning about state-backed groups using it since February 2025 Microsoft .
Then the kit takes a second bite, sending the victim on to a fake Microsoft 365 sign-in page that collects the password and session cookie as well, in case the first route gets cut off.
The Google route is simpler. A pop-up relays Google’s real sign-in through the attacker’s server, which records the password, the MFA answer and the session cookie on the way past (one domain Push caught was continue-to-event.claudewebinnar.com) Push Security .
02 Why “check the link” and MFA fail here
Hover over the button in the email and you see a Google or Zapier address; follow it to the end and the place you type your password is Microsoft’s own domain, with a valid certificate and the familiar logo. Every check you were trained to make comes back clean.
MFA does not help either, because you do the MFA yourself, on the real page, on behalf of the attacker. BleepingComputer put it plainly: with device code sign-in the attacker bypasses MFA entirely BleepingComputer .
The targeting is deliberate. Push says both kits go after marketing staff, and the lures are written for them: a “ChatGPT for Social & Digital Marketing Teams” session BleepingComputer , a SpaceX briefing on AI for marketing, a Claude event for “social, paid media and digital marketing teams”. Marketers are expected to try every new AI tool first, they get event invitations every week, and their accounts often control the ad budget, the company’s social pages and the customer list, which makes them worth more to an attacker than their job title suggests.
The reward is the tell. Nobody pays $500 for ninety minutes in a webinar.
03 What stops it
Two settings close almost all of this, neither is new, and neither is on by default: Microsoft 365 allows device code sign-in until someone blocks it.
The first is a Conditional Access policy that blocks device code sign-in. Microsoft’s own documentation says it recommends blocking device code flow wherever possible, using the “Authentication flows” condition Microsoft Learn . Almost nobody in an office needs it, so exclude the handful of conference-room devices that do and block it for everyone else.
The second is phishing-resistant sign-in with passkeys or hardware security keys. They are tied to the real website’s address, so a proxy sitting in between gets nothing it can reuse, which kills the Google route and the fake Microsoft 365 page (the proxy route in the table below). On its own it does not kill the device code route, because there you really are signing in on the genuine page.
You need both.
| Defence | Code route (Microsoft) | Proxy route (Google) |
|---|---|---|
| Hover over the link | No. Short link, then Microsoft’s real domain | Rarely. The pop-up domain is odd, but few people check a pop-up |
| SMS or push MFA | No. You approve it yourself | No. The proxy relays it and keeps the session |
| Block device code flow | Yes | No |
| Passkeys or security keys | No, on their own | Yes |
| Revoke sessions when someone reports it | Limits the damage | Limits the damage |
| “Find the event on the company’s own site” | Yes, if people do it | Yes, if people do it |
If someone has already entered a code, sign them out everywhere straight away by revoking their sessions, which is what Microsoft recommends when device code phishing is suspected Microsoft , and then check the account for new mail rules, newly registered devices and app consents the attacker may have added while they were in.
Then find out whether the next one would work. Ask your Microsoft 365 admin one question on Monday: is device code sign-in blocked for us? If the answer is “I’d have to check”, our free audit will tell you, along with who in your team would have confirmed the invite.
Questions readers ask
Is the Anthropic event invitation email real?
Treat any unexpected “private” Anthropic, OpenAI or SpaceX event invitation as phishing until you find the same event on the company’s own website, reached by typing the address yourself. Since September 2026 security firms Cofense and Push Security have documented fake invitations under all three brands, offering free subscriptions or tokens and asking you to register with your work Microsoft or Google account. The goal is to take over that account.
Does MFA protect me from fake event phishing?
Not from this campaign. On the Microsoft route the victim types a device code into Microsoft’s genuine sign-in page and completes MFA there, which hands a signed-in session to the attacker. On the Google route a proxy relays the real sign-in and captures the session cookie after MFA. Blocking device code sign-in and using passkeys or security keys are what close these two routes.
How do I block device code phishing in Microsoft 365?
In Microsoft Entra ID, create a Conditional Access policy that uses the “Authentication flows” condition, select device code flow, and set it to block. Exclude only the few accounts or devices that genuinely need it, such as conference-room systems. Microsoft’s own guidance says it recommends blocking device code flow wherever possible.
References
Sources
- Cofense Intelligence. “Would your employees join a fake Anthropic AI preview?” 29 October 2026. cofense.com
- Jennings, L. “AI-themed fake event campaign analysis.” Push Security, 5 November 2026. pushsecurity.com
- Abrams, L. “Fake OpenAI, SpaceX events used to steal Microsoft, Google accounts.” BleepingComputer, 5 November 2026. bleepingcomputer.com
- Microsoft Learn. “Authentication flows as a condition in Conditional Access policy.” Microsoft Entra ID documentation, 2026. learn.microsoft.com
- Microsoft Threat Intelligence. “Storm-2372 conducts device code phishing campaign.” 13 February 2025, updated 31 July 2026. microsoft.com
Close the gap this note describes
- Penetration testing You see the way in an attacker would use today, how far it reaches and what to close first.
- Free audit We map what you expose to the internet and send back the routes in, ranked by what each one reaches.
- Dark web check for partners For IT providers: name one client and we sweep breach dumps and forums for their leaked credentials.