Cybercriminals Plant Malicious AI Agents in Open Source Tool Repositories

ESET analyzed 900,000 AI agent skills and found a rapid increase in suspicious and malicious toolsets—suspicious skills rose from ~10,000 to >25,000 and malicious ones from ~600 to >3,000—capable of browsing the web, executing commands, exfiltrating data, and deploying credential-stealing tools like Mimikatz; the report warns this proliferation of agentic AI tools expands the attack surface and urges organizations to restrict untrusted tools and enforce controls.