Google: Autonomous AI Agents Stole Thousands of Credentials in Under 6 Hours

Threat actors used a multi-agent AI framework to compromise thousands of credentials in under six hours. Google's Threat Intelligence Group (GTIG) traced the campaign to a financially motivated actor that breached an organization's cloud infrastructure, assembling an autonomous framework from an AI chatbot, a prompt, and agent instructions.

The system ran troubleshooting and IP rotation autonomously, routing traffic through the victim's own addresses to appear legitimate. The criminal group UNC6780 (TeamPCP) has run large-scale compromises across PyPI, npm, and Docker Hub, publishing trojanized forks of Model Context Protocol servers. Malware loaders included prompt injections with extreme requests about biological and nuclear weapons, designed to bypass LLM security scanners. "Criminals, like the ones who conducted a mass exploitation campaign in just six hours, will gravitate to attacks that are faster than we can respond to." — John Hultquist