Cyberattacks can now move from spotting a software weakness to attempting to exploit it without a person guiding each step, according to Google’s Threat Intelligence Group (GTIG). The group says it observed attackers fully automating vulnerability discovery and exploitation for the first time.
How the system works
GTIG found an exposed server running an AI framework called “exploit_agent.” The framework connects to multiple language models and automates a chain of tasks that would otherwise require someone to identify a weakness and act on it.
The system compares software code from before and after a patch. By examining what changed, it can identify vulnerabilities, develop exploits and run them automatically. In other words, the same system is described as handling both the search for a software flaw and the attempt to use it.
That combination matters because it links analysis and action in one automated process. The report does not describe a human approving each exploit before it runs; GTIG characterizes the activity as having no human in the loop.
What the evidence says about the operator
The server was exposed, giving GTIG a view of the framework and its configuration. Russian-language comments in configuration files suggest the operator speaks Russian. That is a clue about the operator, rather than a definitive identification.
Server data also suggests the operator targeted French websites in media, entertainment and education. The reported information describes the targets and the system’s capabilities; it does not establish the outcome of each attempted attack.
Those distinctions help keep the finding in perspective. The observation is about an automated approach and evidence on a server. The language in configuration files and the apparent choice of targets offer indications, but they do not by themselves identify who operated the system.
Speed comes with visible traces
GTIG calls the development a critical turning point. At the same time, the group says the attacks currently sacrifice stealth for speed. The automated activity leaves forensic traces that make it easier to detect.
This creates a tension in the system’s design. Automation can carry out vulnerability discovery and exploitation rapidly, but the traces left behind can give investigators evidence to examine. Faster execution does not necessarily mean quieter activity.
For defenders, the finding points to the importance of looking for signs left by automated exploitation. GTIG’s observation suggests that the system’s speed and its detectability are linked: the process can run without a person guiding each step, yet its forensic traces remain visible.
A turning point, with limits
The report describes a specific observed system, not proof that every attacker now uses fully automated cyberattacks. It also does not say that automation has made attacks undetectable. On the contrary, GTIG says the current approach leaves traces that can make it easier to spot.
The shift is still significant because it changes how much of the process can be automated. A system that compares code around a patch, finds vulnerabilities, develops exploits and runs them reduces the need for human involvement across those steps.
GTIG’s assessment therefore carries two ideas at once: automated vulnerability exploitation has arrived in the activity it observed, and the current system’s emphasis on speed leaves evidence behind. The report frames this as an important change while describing a practical limit in the attacks seen so far.