Anthropic reports multiple real-world threat groups used Claude to support cyber operations, including credential harvesting and data theft across many victims. The report includes specific, simulation-ready lures such as a fake ESET NOD32 login portal that sends stolen passwords to Telegram and a ClickFix-style Windows “Run” dialog trick.
Anthropic disclosed that multiple threat groups used its Claude AI models not just for simple queries, but inside multi-agent frameworks capable of performing reconnaissance, exploitation, and data theft with minimal human supervision. Two specific operations stand out: an Iran-nexus actor (tracked as GTG-30006) that built phishing infrastructure including a bogus ESET NOD32 antivirus login page, and a Russian/Ukrainian-speaking group (GTG-50021) that ran a fraudulent AI reseller scheme targeting people trying to buy discounted Claude access.
The ESET-themed lure directed victims to a fake antivirus login portal. Once a victim entered their username and password, the credentials were transmitted to Telegram rather than to any legitimate ESET system. A second lure used a ClickFix-style technique, instructing victims to open the Windows Run dialog and execute a command shown on an untrusted web page, a pattern increasingly used to get users to run malicious commands themselves.
Separately, the fraudulent AI reseller operation offered cheap access to Claude outside of official channels. Buyers who logged in with their Anthropic account credentials to activate this discounted access instead had a credential harvester installed, which siphoned their account credentials for resale.
Each scenario relied on a familiar trust shortcut:
The report also notes these operations ran largely autonomously through multi-agent frameworks, which can increase the speed and volume of attempts an organization may face.
Organizations should reinforce simple, consistent verification habits: treat any prompt to re-enter credentials for security tools as suspicious, and confirm through a known-good route such as an official app or internal IT portal. Employees should be told explicitly never to run commands from a website's instructions, and to report such pages to security. Procurement and developer teams should only acquire AI tools through approved vendors to avoid reseller schemes that harvest credentials. Because AI-assisted operations can scale quickly with little human oversight, awareness training should emphasize that verification steps matter even when an attack looks automated or high-volume.
Mirage safely runs attacks like this one against your own team, so you find out what happens before a real adversary does.
Anthropic reported that multiple threat actors used Claude in multi-agent frameworks that performed reconnaissance, exploitation, and data theft, going beyond simple question and answer use.
An Iran-nexus actor built a bogus ESET NOD32 antivirus login page that captured credentials and transmitted them to Telegram.
It is a lure that mimics a fix instruction, prompting a user to open the Windows Run dialog and execute attacker-provided commands.
A Russian/Ukrainian-speaking group ran a fraudulent AI reseller operation offering cheap Claude access, then installed a credential harvester to steal Anthropic account credentials.
Attackers are now using AI like Claude to build pro-level phishing kits at scale. One real case: a fake ESET NOD32 antivirus login page that looked legit, but every password typed there was quietly sent to Telegram. Another: a ClickFix-style web page telling users to open the Windows Run box and paste in a command as a 'quick fix', that’s how malware gets in. Aha moment: if a page wants you to re-login to security software or run commands from Windows Run, stop. Don’t do it, screenshot it and report it to IT immediately.
This news roundup describes real social-engineering operations where attackers impersonate trusted support teams to trick people into giving up secrets. One campaign uses email/SMS/WhatsApp plus AI voice calls pretending to be Apple Support to steal iPhone passcodes, while another uses phishing…
This threat trend report describes multiple real-world APT campaigns where attackers rely on social engineering and trusted services (Zoom/Teams, Telegram, webmail, GitHub) to steal credentials and access cloud accounts. Notable examples include fake meeting lures to deliver malware, and abuse of…
This recap describes multiple real-world social-engineering-driven attacks, including vishing calls that push employees to spoofed login pages and a supply-chain trick where cloning/opening a GitHub repo in developer tools triggers malware. It also highlights an unusual case where an AI model…
North Korea-linked BlueNoroff is using fake Zoom and Microsoft Teams meeting pages shared from hijacked trusted contacts (often via Telegram) to trick victims into “updating” Zoom/Teams and running malicious commands. The phishing kit also fingerprints the victim’s browser to identify installed…
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…
Microsoft described two real-world campaigns: an invoice fraud blast impersonating executives to trick finance teams into ACH payments, and a passkey-themed helpdesk scam that steals or bypasses authentication to take over Microsoft cloud accounts. In the second campaign, victims are called or…
