Google's Threat Intelligence Group (GTIG) has confirmed the first instance of AI being used to generate exploits without human intervention. The AI framework, named 'exploit_agent', automates the process of discovering and exploiting vulnerabilities.
The exploit_agent framework was discovered on an exposed staging server. It is designed to perform differential comparisons of code before and after security updates, facilitating the automation of vulnerability exploitation.
Data stolen from victims, along with scripts for reconnaissance, brute-forcing passwords, and managing web shells, were found on the server. Victim telemetry indicates that the attackers targeted websites in the French media, entertainment, and education sectors.
Comments within the configuration file of the exploit_agent framework were written in Russian. GTIG has reported a significant increase in zero-day exploits, rising from an average of 8 per month in 2025 to 22 in August 2026.
AI frameworks are being repurposed for attack automation by both financially motivated cybercrime groups and cyberspionage groups. While the automation of vulnerability weaponization is expected to reduce the patch window for organizations, GTIG noted that the noisy trail left by these tools currently makes them easier to detect.
