MOUNTAIN VIEW, Calif. - Google's Threat Intelligence Group reported on Nov. 16, 2026, its first observation of hackers using an autonomous artificial intelligence framework to turn software security patches into live exploits without human intervention.
Autonomous Framework Compares Code Around Security Updates
Researchers discovered a multi-agent system named exploit_agent on an exposed staging server alongside stolen victim data and post-breach scripts. Configured with Russian-language comments and targeting French media, entertainment, and education sites, the tool connects to large language models to compare web software before and after patches, pinpoint fixed flaws, and automatically launch attacks.
Zero-Day Exploitation Climbs Alongside Open-Source Attack Kits
Google's quarterly AI Threat Tracker showed monthly zero-day exploits rose from an average of eight in 2025 to 11 in 2026, peaking at 22 in August. Criminals also stole cloud AI credentials causing single-day fraudulent bills exceeding $100,000, while open-source offensive agents like CAIRN and PentAGI targeted U.S. aerospace firms and Asian organizations, though Google noted these frameworks trade stealth for velocity.
Shorter Patch Windows Demand Rapid Software Updates
Because automated patch comparison shrinks the gap between fix release and exploitation, Google advised organizations to update internet-facing software within days rather than weeks. In Hackread's summary of the Nov. 16 findings, researchers emphasized that Google Threat Intelligence Group says it has observed a threat actor using an AI framework to identify software vulnerabilities, write exploits and execute them without a person directing each step.



