Malicious OpenClaw Skills Used to Distribute Atomic MacOS Stealer

Source: Dev.to

Overview

Atomic Stealer (AMOS) has evolved from its traditional distribution via cracked software to a sophisticated supply‑chain attack targeting AI‑agentic workflows. Attackers embed malicious instructions in SKILL.md files on platforms such as OpenClaw, manipulating AI agents to act as trusted intermediaries. By deceiving the AI into presenting fake setup requirements, the malware tricks users into manually facilitating infection on macOS systems.

Technical Details

This variant employs Mach‑O universal binaries and multi‑key XOR encryption to evade detection while harvesting a broad range of sensitive data. Targets include Apple and KeePass keychains, browser credentials, cryptocurrency wallets, and private messages. Although it lacks typical persistence mechanisms, the use of encrypted universal binaries makes static analysis more difficult.

Impact

The ability to exploit the trust relationship between users and AI agents represents a significant shift in social‑engineering tactics within the cybersecurity landscape. Even without persistent footholds, the attack’s reliance on user‑mediated execution through AI‑driven instructions amplifies its potential reach and effectiveness.