Google Threat Intelligence Group (GTIG) observed threat actors transitioning from simple AI coding assistants to sophisticated multi-agent AI frameworks that automate entire attack lifecycles. These AI agents perform vulnerability scanning, credential harvesting, troubleshooting, IP rotation, and traffic routing through compromised cloud environments to evade detection. Examples include a mass credential-harvesting campaign deployed within six hours and an exposed command-and-control server managing over 23,800 stolen secrets. State-backed groups also use AI for reconnaissance, phishing, malware development, and exploitation. Despite these advances, fully autonomous zero-day discovery and exploitation pipelines have not been observed in real-world attacks. Google’s AI model Gemini detected and mitigated many such abuses early.

The use of AI-driven multi-agent frameworks enables attackers to conduct large-scale credential theft and automate complex attack workflows with reduced human involvement, increasing attack speed and scale. This reduces defenders' response windows and complicates detection efforts. The automation of credential management and reconnaissance enhances attackers' operational capabilities. However, fully autonomous exploitation of zero-day vulnerabilities has not been observed, indicating current limitations in AI-driven attacks.

Google’s AI defenses, including the Gemini model, have detected and disrupted many AI-powered attack campaigns, and associated accounts have been banned. Organizations should monitor vendor advisories for updates on AI-related threat detection capabilities. No specific patches or fixes apply as this is a threat actor technique rather than a software vulnerability. Defensive measures should focus on credential protection, anomaly detection, and rapid incident response to AI-augmented attacks.