Experts warn malicious AI skills are hitting more victims than ever — with one family amassing 1.7 million downloads

Security researchers at Zenity Labs discovered a large-scale credential-stealing campaign exploiting AI skills on skills.sh, with one malicious skill family reaching over 1.7 million installs. Attackers cloned legitimate AI skills, later injecting code to exfiltrate sensitive credentials and metadata. Vercel and Microsoft removed the malicious skills after disclosure, but users must manually uninstall them to ensure safety.