APT Group Deploys Fake Claude App in APAC

SilverFox has begun distributing counterfeit versions of Anthropic’s Claude chatbot to breach enterprises across the Asia‑Pacific region, according to a recent briefing by Kaspersky’s Global Research and Analysis Team (GReAT). The group’s latest tactics exploit the growing reliance on large‑language‑model assistants in corporate workflows.

Fake Claude apps serve as new infection vector

Researchers observed that the threat actor packages malicious payloads within installers for Windows, macOS and Linux that masquerade as the official Claude client. When users run the fake installer, a stealthy backdoor is dropped, allowing long‑term espionage and data exfiltration. The campaign targets firms in India, Indonesia, South Africa and Russia, spanning industrial, consulting, trade and transportation sectors.

Phishing emails accompanying the fake apps mimic tax‑audit notices or claim to contain a “list of tax violations.” The messages urge recipients to download a compressed file, leveraging the perceived authority of tax agencies. Between January and February 2026, more than 1,600 such emails were recorded.

The distribution chain uses a segmented infrastructure: separate domains and IP blocks host the initial lure, the installer, and the command‑and‑control servers. This design limits the chance that blocking one element will disrupt the entire operation.

Regional focus and sector preferences

Analysis of recent activity shows the group concentrates its efforts in the Greater China area, where over 90 % of attacks originated. Mainland China alone accounts for roughly 71 %. Smaller but notable volumes were seen in Myanmar, Cambodia and Singapore.

Manufacturing receives the highest share of attacks, representing more than a third of incidents. IT services and consulting follow closely, with healthcare and finance also appearing on the list of high‑value targets.

In practice, a mid‑level manager who receives a seemingly legitimate Claude update may unwittingly install the malware, granting the attackers persistent access to internal documents and network traffic. This lowers the barrier for espionage, especially for companies that have recently adopted AI‑assisted tools without robust vetting processes.

AI‑driven ransomware and new attack methods

GReAT also highlighted the emergence of “JADEPUFFER,” the first ransomware fully powered by a large‑language‑model. Unlike earlier strains that required human operators to guide each step, this malware autonomously diagnosed failed attempts and adjusted tactics.

Defensive recommendations

Experts advise organizations to adopt proactive threat hunting powered by artificial intelligence, implement strict Zero Trust controls, and maintain layered defenses across endpoints, networks, applications and data stores. Using AI against AI—deploying large models for real‑time detection—can help keep pace with rapidly evolving threats.

“When attackers can leverage artificial intelligence to automate decision‑making and accelerate every stage of an attack, defenders must respond with the same level of intelligence,” the lead researcher said.