Google finds AI agents writing and launching exploits with no human involved

Google finds AI agents writing and launching exploits with no human involved. This next one is a genuine inflection point.

This is a DailyCyber.News update, brought to you by BareMetalCyber.com, for Tuesday, November 17th, 2026.

This next one is a genuine inflection point. Google's Threat Intelligence Group and Mandiant say threat actors have, for the first time, used AI agents to fully automate vulnerability discovery and exploitation with no human in the loop. Researchers found an exposed staging server running a tool called exploit underscore agent that compared code before and after patch releases, drafted its own exploits, and launched them automatically, with victim data suggesting interest in French media and education websites. Separately, the FBI warned that a framework called ARTEX is being used by Chinese state-linked and financially motivated groups against financial services, defense, and healthcare organizations. And Barracuda confirmed it got hit by an ARTEX-powered intrusion after attackers stole a cloud access key that had been issued to a third-party vendor and never retired. The attacker touched about ten cloud accounts over forty-four hours. Here's the thing though: these tools still succeed mostly because of basic gaps, unrotated credentials, exposed keys, unpatched software. If you're a leader, assume the time between a patch release and a working attack is now measured in hours, not days. If you're a defender, the fundamentals still matter most: rotate vendor keys the moment a project ends, and tighten your patch deployment speed for anything internet-facing.

For the sources and the full Daily Cyber newsletter, visit DailyCyber.News.

Google finds AI agents writing and launching exploits with no human involved
Broadcast by