Security News

Google Reports First Fully Automated AI Cyberattack, With No Human Directing the Intrusion

CyberScoop · 17 Nov 2026
Key Takeaway Treat vendor security patches as urgent, because attackers can now turn a published fix into a working exploit within days.

Google has reported what its analysts believe is the first cyberattack run from start to finish by artificial intelligence, with no human involved at any stage. The finding appears in a quarterly report from Google's Threat Intelligence Group and its Mandiant incident response unit. Researchers did not name a group or state, but pointed to Russian-language comments in the operation's code. Victim data on the staging server suggested the targets were French websites in the media, entertainment and education sectors.

At the centre of the operation was a system the researchers called "exploit_agent". It links multiple AI agents, using both frontier and locally hosted language models, to compare widely used software before and after vendors release patches. When a fix appears, the system works out what the change reveals about the flaw, drafts an exploit and runs it automatically. Investigators also found scripts for internal reconnaissance, password guessing and installing webshells, including one that removed the webshells of rival attackers.

Until now, Google's trackers mostly saw attackers using AI to speed up work that people still directed, such as writing phishing lures, producing code and summarising stolen data. Google calls the new report a critical inflection point, because the path from vulnerability research to intrusion is now running unattended. It also saw open-source offensive frameworks, including CAIRN and PentAGI, used against a US aerospace organisation and organisations across Asia, and cited a China-nexus group building an automated attack framework for likely espionage.

Speed is the clear advantage. After vendors patched two PHP vulnerabilities on 21 October, hundreds of purported exploits appeared within days, most nonfunctional and likely AI-generated. Google counted 22 exploited zero-days in August alone, up from 11 per month earlier in 2026 and an average of eight in 2025.

Why it matters

Many small businesses assume they have a grace period after a patch is released, because attackers need time to reverse-engineer the fix. Automated patch comparison and exploit drafting erodes that window, so slow patch cycles on internet-facing software become the main exposure. The rival-webshell clean-up script also suggests compromised servers may be contested by several attackers.

What to do

  • Prioritise patching internet-facing software, including PHP-based websites, soon after vendors release fixes.
  • Review web servers for unexpected webshells and unfamiliar files.
  • Check logs for repeated password guessing against exposed logins.
  • Enable automatic updates where the vendor supports them.
AI cyberattack Google Threat Intelligence Mandiant patch management zero-day automation

Summarised by CISO AI from CyberScoop, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.