Malicious OpenClaw Skills Used to Trick Users into Manual Password Entry for AMOS Infection


Atomic macOS Stealer (AMOS), a well-known data-theft malware, has taken a sharp turn in how it reaches victims.

Instead of hiding inside cracked software downloads as it once did, threat actors now embed it within malicious OpenClaw skills — small add-on packages that extend AI agent capabilities on platforms like OpenClaw.

AMOS operates as a malware-as-a-service (MaaS) tool built to steal sensitive data from Apple users.

It harvests a wide range of information, including credentials, browser data, cryptocurrency wallet details, Telegram chats, VPN profiles, Apple keychain items, and files from common folders such as Desktop, Documents, and Downloads.

Trend Micro analysts identified a new AMOS variant embedded in OpenClaw skills and tracked the campaign across multiple repositories.

Threat actors had uploaded 39 malicious skills to ClawHub, SkillsMP, and GitHub, with over 2,200 malicious skills eventually discovered on GitHub alone.

google

This campaign marks a clear shift from earlier AMOS delivery methods and introduces a new form of supply chain attack targeting AI agent workflows.

The attack begins with a SKILL.md file that looks completely harmless. It tells the AI agent to install a fake prerequisite called “OpenClawCLI” from a malicious external website.

GPT-4o constantly reminding the user to manually install the malicious 'driver' (Source - Trend Micro)
GPT-4o constantly reminding the user to manually install the malicious ‘driver’ (Source – Trend Micro)

When a less cautious model like GPT-4o processes the instruction, it either installs the tool silently or continuously prompts the user to install the fake “driver” manually.

Claude-4.5-Opus identifies the skill as malicious (Source - Trend Micro)
Claude-4.5-Opus identifies the skill as malicious (Source – Trend Micro)

More capable models like Claude Opus 4.5 flag the skill as suspicious and refuse to proceed.

If the user or AI agent presses ahead, a Base64-encoded command is fetched and executed, dropping a Mach-O universal binary that runs on both Intel-based and Apple Silicon Mac machines.

Fake dialogue box asking users for their password (Source - Trend Micro)
Fake dialogue box asking users for their password (Source – Trend Micro)

When macOS rejects the unsigned file, a fake password dialogue box appears, tricking the user into entering their system password — giving the malware exactly the access it needs to proceed.

Inside the Infection Chain

Once the password is entered, AMOS begins collecting data immediately.

It gathers the machine’s username and password, files from Desktop, Downloads, and Documents folders (including .pdf, .csv, .kdbx, and .docx formats), Apple keychain credentials, and Apple Notes.

The malware also targets 19 browsers for stored cookies, passwords, and credit card data, and can reach across 150 cryptocurrency wallets.

AMOS Stealer process accessing sensitive data on the affected macOS system (Source - Trend Micro)
AMOS Stealer process accessing sensitive data on the affected macOS system (Source – Trend Micro)

All collected data is compressed into a ZIP archive and uploaded to a command-and-control (C&C) server at socifiapp[.]com.

Users are advised to verify the source of any OpenClaw skill before running it, avoid entering system passwords prompted by unfamiliar tools, test unvalidated skills in an isolated environment, and use containers to limit AI agent execution.

IoCs

TypeIndicatorDescription
URLhxxps://openclawcli[.]vercel[.]app/Malicious skill delivery site
IP Address91.92.242[.]30Payload download server
URLhxxp://91.92.242[.]30/ece0f208u7uqhs6xPayload download URL
File Nameil24xgriequcys45Mach-O universal binary (AMOS payload)
C2 Serversocifiapp[.]comCommand-and-control exfiltration endpoint
Detection NameTrojan.MacOS.AmosAMOS malware detection name

Follow us on Google News, LinkedIn, and X to Get More Instant UpdatesSet CSN as a Preferred Source in Google.

googlenews



Source link