Google finds an attacker tool that hunts and exploits flaws with no human
Google's Threat Intelligence Group (GTIG) reports that attackers have, for the first time it has observed, automated both finding software vulnerabilities and exploiting them, with no human involved. The finding appears in the group's AI Threat Tracker.
Investigators came across an exposed server running an AI framework called "exploit_agent". It connects to several language models. It compares software code before and after a patch, identifies the vulnerability the patch fixed, builds an exploit and runs it automatically. Google rates the activity as Level 5, meaning AI tools carry out multiple attack stages with minimal human involvement, and calls the moment a critical turning point.
Why it matters: the scarce resource in offense has been skilled human time. Turning a patch into a working exploit used to take an expert; a loop of models and scripts makes that work cheap and repeatable. That shifts the economics toward attackers and shortens the window between a fix being published and systems being hit. The services people depend on, from media to education, sit on that exposed ground. The same model-driven analysis is also available to defenders, which is where the knowledge side of this story lies.
What to watch: Google says the automated attacks currently trade stealth for speed, leaving forensic traces that make them easier to detect, but it forecasts that more powerful attacks are only months away. That timeline is Google's own prediction. Configuration files carry Russian-language comments, which only suggests a Russian-speaking operator, and server data points to French media, entertainment and education sites as targets. Neither is confirmed.
Source: The Decoder
Many Minded