Malicious AI Skills Campaign Targets 1.7 Million Users, Exposes Major Vulnerabilities

Zenity Labs revealed a malicious skills campaign that has affected over 1.7 million installs through Vercel's skills.sh. The campaign involved credential-stealing skills that were initially benign but later weaponized. Attackers inserted malicious installation instructions to download and execute harmful code on user systems. The malware targets sensitive data, including SSH keys and cloud credentials. Zenity also identified dozens of additional dangerous skills that manipulate AI agents to download and execute files from attacker-controlled endpoints. Major AI platforms like Microsoft, Google, and OpenAI were found to have vulnerabilities that could lead to data theft and operational disruption. Zenity Labs disclosed these findings at Black Hat USA, prompting immediate patches from some affected companies.

Key Points: • Malicious skills campaign affected over 1.7 million installs, targeting AI tools. • Attackers used typosquatting and benign initial skills to execute harmful code. • Major AI platforms have vulnerabilities that allow for data theft and manipulation.