Cybercriminals Plant Malicious AI Agents in Open Source Tool Repositories (infosecurity-magazine.com)

• ESET analyzed 900,000 AI skills; found thousands malicious
• Suspicious skills rose from 10,000 to 25,000; malicious from 600 to 3,000
• Malicious AI agents can exfiltrate data, drop malware, hijack commands

"ESET researchers found thousands of malicious AI agents in public repositories, capable of data theft, malware execution, and command manipulation. Suspicious skills grew from 10,000 to over 25,000, and malicious from 600 to over 3,000. Some tools disguised as legitimate red-teaming tools exfiltrated credentials and dropped Mimikatz. Organizations are advised to restrict suspicious AI tools and educate users."