The email flatters you first. It praises your research, calls the event important, and offers a “VIP seat” if you register today. Nothing looks sloppy. That’s the whole problem.
Cisco Talos, Cisco’s security research group, published findings on October 8 about a campaign it tracks as UAT-11985. Two write-ups I read, one from Gridinsoft summarizing the research and one from GBHackers on October 12, say the attackers sent tailored invitations to people tied to research organizations in Taiwan. The goal wasn’t to infect a laptop. It was to take over a Google account, even with two-factor authentication turned on.
I couldn’t get to the original Talos report this morning, so everything here comes from those two write-ups. They agree with each other, but treat the details as reported, not independently confirmed.
How the lure worked
The invitations posed as three real institutions: the Taiwan European Union Centre, the NCCU Institute of International Relations, and the Taiwan Research Institute. The attackers reused real, public event details to look credible.
Every message ran the same way: a short policy introduction, some personal praise, the “VIP seat,” then registration details. According to Gridinsoft, an invitee who checked with the organizations found that none of them could confirm the three named senders worked there.
The registration links were labeled as Google Forms, but the label hid a destination the attackers controlled. In some cases they also copied real event posters and swapped the QR code. Talos put an original poster next to a modified copy: same familiar layout, different destination. It suggested the attackers may have hoped posters would be displayed where colleagues could scan them, though that is a possible route, not a confirmed one.
Where the two-factor login breaks
The registration page led to a fake Google sign-in page. This is where it gets interesting.
The kit is what’s known as an adversary-in-the-middle framework. It passes whatever you type to the real Google in real time. According to the write-ups, a persistent WebSocket connection let the operator control which screen the victim saw. The backend sent an account identifier to Google and submitted the captured password right away. When Google asked for another verification step, the operator could show the victim a matching prompt.
Put simply: you type your password into the fake page, and the attacker types it into the real one. Google sends a challenge, and the fake page shows you the same challenge. You complete it. The attacker grabs the signed-in session and walks in.
GBHackers stresses that this is a phishing technique, not a flaw in Google’s systems. The kit even ends on a locally hosted “success” page, so a familiar finish screen proves nothing.
The AI question
Talos judged that the emails showed signs of AI help, based on the language and the personalization, but said it could not confirm that. The caution is worth keeping. “AI wrote it” is an easy headline and a hard thing to prove.
Here’s what is easier to say. Tools that write fluent, personalized text have weakened one old warning sign. For years, people were told to watch for clumsy grammar. I use AI tools day to day, and they turn out clean, polite, well-organized prose in seconds, in any tone I ask for. Polish is no longer evidence of legitimacy. A flattering, well-written email tells you almost nothing about who sent it.
What the researchers recommend
The defenses in the Gridinsoft summary are practical, and none of them is exotic:
- Open the organization’s event page yourself, not through the email, and confirm the organizer with contact details you find there.
- Check the real link destination, or the QR preview, before you enter any account information.
- Entered credentials, or approved a prompt you didn’t expect? Follow Google’s compromised-account steps. Review security events and devices, secure your password and recovery information, and check Gmail forwarding and other settings for changes you didn’t make.
- On an account managed by an employer or school, tell the administrator.
The same write-up notes that Google describes passkeys as phishing resistant, because they are tied to the site or app they were created for. That’s Google’s claim, relayed secondhand here, but it fits how the technology is designed. A passkey won’t hand over a sign-in to a lookalike domain, because the domain doesn’t match.
My take: watch the address bar, not the email
The usual advice is to scrutinize the message. I think that’s the wrong place to spend your attention now. If the writing can be perfect, the personalization accurate, and the poster a pixel-for-pixel copy, the message isn’t where the signal lives anymore.
The signal lives in where you’re about to type your password. A sign-in that starts from a link or QR code you didn’t choose deserves suspicion, however good the invitation looks. Open the service yourself, or use a passkey, which does that check for you.
None of this means two-factor authentication is useless. It still blocks plenty of simple password theft. It just can’t protect you from a page that sits in the middle and relays everything.
What to do this week
Start with a passkey, if your Google account supports them. Then make it a habit never to sign in from an invitation, a QR poster, or a “registration required” link; go to the site directly. If you manage a team, share this campaign as an example. The lure was aimed at researchers and policy staff, not at obvious targets.
The campaign Talos described was narrow, aimed at one community. The technique isn’t. Expect the same relay approach, and the same flattering tone, to show up wherever a convincing invitation can be written cheaply, whether that’s a conference, a job offer or a shared document. The open question for the next year is whether our sign-in habits change faster than the lures improve.
