Anthropic says it disrupted a Russian state-linked campaign that used Claude to continuously rebuild malware when security tools detected it. The group (GTG-20006, linked to Midnight Blizzard/APT29) ran phishing and other human-targeted schemes, including device-code token theft against Microsoft 365 and fake update lures, to compromise government and diplomatic targets, often staging malware on disposable servers.
A Russian state-linked group tracked as GTG-20006, associated with Midnight Blizzard/APT29, ran a multi-stage social engineering campaign against government, diplomatic, and defense-related targets. The operation combined a device-code phishing framework called Embassy Kit to steal Microsoft 365 authentication tokens, fake software update prompts to deliver Windows credential stealers, and compromised hotel Wi-Fi infrastructure to redirect traveling officials to malicious ClickFix-style lures. Malware was staged on disposable hosting servers, and victims were funneled to these servers through phishing, ClickFix instructions, and DNS hijacking after attackers modified DNS records using compromised admin credentials.
What sets this campaign apart is the reported use of AI to sustain the operation. Monitoring agents tracked whether deployed malware was flagged by security products, and when detection occurred, the agents autonomously modified and rebuilt the malware to bypass those defenses. This effectively automated a cat-and-mouse cycle that traditionally required manual developer effort, letting the attackers iterate faster than defenders could update signatures.
Each lure exploited a moment of routine trust: authenticating to Microsoft 365, applying a system update, or reconnecting to hotel Wi-Fi while traveling. None of these actions are inherently suspicious, which is exactly why they work as pretexts. The device-code flow in particular abuses a legitimate Microsoft sign-in mechanism, making it harder for a target to distinguish a real request from a malicious one. The hotel Wi-Fi angle also targeted travelers, who tend to be more rushed and less attentive to security prompts than they would be at their normal workstation.
Organizations with staff who travel or handle sensitive government, diplomatic, or defense work should reinforce that device-code sign-ins and software updates must go through verified, known channels only. Encourage rapid reporting of unusual prompts, since the underlying malware and lures in this campaign were designed to be rebuilt and changed quickly once detected, meaning static indicators alone will not keep pace with the threat.
Mirage safely runs attacks like this one against your own team, so you find out what happens before a real adversary does.
Device-code phishing tricks a user into entering a code on a legitimate Microsoft 365 sign-in page, which then grants the attacker an authenticated session token without needing a password. In this campaign it was used via a framework called Embassy Kit to steal Microsoft 365 tokens from diplomatic and government staff.
The group compromised hotel guest Wi-Fi vendors and modified DNS records to redirect traveling victims, then served ClickFix-style 'fix your connection' instructions that delivered malware tailored to the victim's Windows, Android, or iOS device.
According to the findings, monitoring agents used Claude to automatically detect when security products flagged their malware and then autonomously modified and rebuilt it to bypass those detections.
Be wary of unexpected device-code sign-in requests, pop-up prompts to install software updates, and unusual captive-portal instructions on hotel or public Wi-Fi, all of which were used as lures in this campaign.
Imagine this email: “Action required: Verify your Microsoft 365 access using this device code.” Looks normal, right? Behind that is GTG-20006, linked to Midnight Blizzard, using an AI toolkit called Embassy Kit to steal Microsoft 365 tokens. You enter the code, they quietly siphon your mail, just like the eight orgs they already hit. Here’s the trap: the email pushes a device code and link that don’t match our normal sign-in flow. Or you’re on hotel Wi‑Fi and a random page pops up: “Critical update required” with a download, actually a Windows credential stealer. If you ever get a device-code email or pop-up you weren’t expecting, stop. Don’t enter the code. Don’t install the “update.” Contact IT through Teams or our helpdesk portal and ask them to confirm it first.
Google says multiple suspected Russia-linked espionage clusters targeted academics, government, and defense-related personnel by abusing legitimate sign-in features instead of using obvious fake login pages. The campaigns used realistic lures (file sharing, conference invites, and “secure WhatsApp”…
Google says three suspected Russian espionage clusters are targeting academics, think tanks, diplomats, and related nonprofit staff by abusing legitimate login and verification workflows that may not look like “classic phishing.” The campaigns include app-password scams, OAuth/device-code tricks,…
Google tracked three suspected Russia-linked groups running targeted phishing that abuses real login and authentication features (app passwords, OAuth, and device codes) to get into accounts. The lures often look like legitimate conference or diplomatic invitations, and some campaigns spoof…
Google says three suspected Russian cyber-espionage groups are running highly targeted phishing campaigns against people in government, academia, defense, and think tanks in the US and Europe. A key theme is abusing legitimate Google/Microsoft OAuth login flows so the outreach looks real, tricking…
Researchers showed how a malicious web page could trick OpenAI’s Atlas AI-enabled browser into taking actions a user didn’t intend, like spamming WhatsApp contacts or modifying an Amazon account. The attacks used prompt-injection style instructions hidden in a seemingly legitimate “newsletter…
A SecurityWeek roundup highlights multiple real-world scams and campaigns where attackers trick people rather than “hack” systems directly. Notable items include OAuth “consent phishing” (getting users to approve a malicious app’s access), and phishing-evasion using invisible Unicode characters…
