Google warns autonomous cyberattacks have arrived
Google's Threat Intelligence Group has documented the first fully automated cyberattack framework that discovers and exploits web vulnerabilities without human intervention.
In its Q4 2026 AI Threat Tracker published November 16, Google's Threat Intelligence Group (GTIG) alongside Mandiant revealed what they call the first fully automated vulnerability discovery and exploitation carried out by malicious actors. The era of autonomous cyberattacks is no longer theoretical but has begun.
The exploit_agent framework
The discovery centers on a compromised staging server where investigators found a multi-agent framework named exploit_agent. This system autonomously connects to large language models to study code of popular web applications, both before and after security patches are applied.
What GTIG flags as new is that exploit_agent runs the whole loop by itself, continuously, spotting weaknesses and writing working exploits without a human supervising.
Attack targets and sophistication
The campaign primarily targeted French websites in the media, entertainment, and education sectors. Investigators found stolen data and post-compromise scripts including brute-force tools and web management systems, confirming the framework was actively exploited rather than tested.
GTIG describes this as an advanced stage of attack automation, representing a significant shift from earlier in 2026 when AI primarily played a supporting role in malicious operations rather than running the entire process autonomously.
Future outlook
The report expects unsophisticated autonomous attacks to appear within one to three months, with more sophisticated versions following within three to six months. GTIG concludes that defenses will need to become equally advanced and automated to counter these evolving threats.