This threat involves a malicious skill named "DeepSeek-Claw" weaponizing the OpenClaw AI agent framework to distribute two malware families: Remcos RAT on Windows and GhostLoader on macOS/Linux. The Windows infection chain uses a signed GoToMeeting MSI sideloading a malicious DLL that patches ETW and AMSI to evade detection, decrypts Remcos RAT with TEA encryption, and enables remote access with capabilities such as keylogging and cookie theft. The macOS/Linux path uses obfuscated Node.js scripts to prompt for fake sudo credentials, harvesting sensitive data including SSH keys, cryptocurrency wallets, and cloud API tokens. The campaign exploits trust in AI agent frameworks and open-source skills to trick developers and AI workflows into executing malicious payloads. Indicators include multiple malicious domains, URLs, and file hashes. There is no known exploit in the wild beyond this campaign, and no vendor advisory or patch is currently available.

The impact includes remote access and control of infected Windows systems via Remcos RAT, enabling data theft such as keylogging and cookie stealing. On macOS and Linux, GhostLoader harvests credentials and sensitive information including SSH keys, cryptocurrency wallets, and cloud API tokens, potentially leading to further compromise and data exfiltration. The campaign exploits AI agent frameworks and developer trust, representing a novel attack vector that could facilitate widespread compromise if the malicious skill is widely adopted.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until official guidance is available, avoid installing or executing untrusted OpenClaw skills, especially those named or related to "DeepSeek-Claw." Monitor for indicators of compromise such as connections to the listed malicious domains and hashes. Employ endpoint detection solutions capable of detecting DLL sideloading, ETW and AMSI patching, and suspicious PowerShell activity. Exercise caution with AI agent frameworks and open-source skills, validating their provenance before use.