Google Threat Intelligence Group says it has identified attackers using AI frameworks to find and exploit software vulnerabilities without a human in the loop. It is the first time the group has observed that level of automation in this activity. Google and Mandiant also report that attackers have used open source agent frameworks during live intrusions.
In one case, the group found an exposed staging server running a framework called exploit_agent. According to Google, it compared web application code before and after patches, identified possible flaws, drafted exploits and ran them automatically. The server also held scripts for activity after a break-in. Data on the server suggested its operator was interested in French media, entertainment and education websites.
Attackers use agent frameworks in intrusions
Google says it independently observed the open source frameworks CAIRN and PentAGI deployed in the wild. It saw CAIRN used against targets that included the aerospace sector in the United States. Between late August and early October 2026, it tracked an actor using a customized platform based on PentAGI to scan organizations across Asia. The group also saw experimentation with CyberStrikeAI and HexStrike AI.
These frameworks can combine security tools and AI models to carry out successive attack tasks. Google says the activity still relies on familiar weaknesses, including compromised credentials and unpatched systems. It also says current autonomous tools leave traffic and code traces that can help defenders detect them.

Patch analysis speeds up exploit attempts
Google reports that attackers are using AI to compare software versions around patch releases and turn findings into exploit attempts. Its analysis found that vulnerability disclosures and exploits nearly doubled between January and August 2026, while monthly zero-day exploits rose from a 2025 average of eight to 11 in 2026 before reaching 22 in August.
Google Security Operations has deployed detection rules for ARTEX, another offensive framework discussed in the report. Mandiant Consulting has released a guide for defending against AI-enabled offensive tools.



