Trusence Every claim has a source
Last updated 17 November 2026 Search Türkçe
Security Developing

Google warns fully automated AI-driven attacks are close, shrinking time to patch

Attackers are already using multi-agent AI to find and exploit flaws, leaving defenders less time to respond.

Updated with 1 new report since it was first published · latest 16 November at 21:06

Google's Threat Intelligence Group reports that attackers have begun using open-source agentic pentesting frameworks such as ARTEX, CAIRN, CyberStrikeAI and HexStrike AI to automate vulnerability discovery and exploitation at machine speed. In one case, a multi-agent system called exploit_agent was found on an exposed staging server and used to attack French media, entertainment and education websites, with the attacker's staging server also holding stolen victim data and webshells. Google also says LLMs are being used to reverse engineer patches through differential analysis, shrinking the window between patch disclosure and exploitation, as seen with PaperCut NG/MF flaws in September. Google expects unsophisticated AI-driven attacks needing little human input within three months and fully autonomous agentic attacks within six. For defenders, the flood of mostly AI-generated proof-of-concept code after PHP2Shell disclosures adds verification work and shortens the time available to patch.

Why it matters

Google's warning suggests the gap between a vulnerability becoming public and it being exploited is narrowing. Attacks that need little human input are expected within three months and fully autonomous ones within six, according to Google. For people who run and maintain internet-facing systems, the practical effect is less time to apply patches before exploitation begins. Google also reports attackers are reverse engineering patches with LLMs and that hundreds of mostly AI-generated proof-of-concept exploits appeared after PHP2Shell was disclosed, which adds to the noise defenders must sort through.

Sources · 2