Loading...
AI hacking now possible without human help, says Google
The revelation was made in a report by the Google Threat Intelligence Group

AI hacking now possible without human help, says Google

Nov 17, 2026
01:06 pm

What's the story

Google's threat researchers have reported the first case of an artificial intelligence (AI) framework being used for hacking, without any human involvement. The revelation was made in a report by the Google Threat Intelligence Group (GTIG), titled "Agentic Threat Activity Crosses a Significant Threshold." The report highlights that the time between a security patch release and its exploitation by attackers is decreasing rapidly.

Automation

exploit_agent automatically drafts and executes exploits

The AI framework, dubbed exploit_agent, was discovered on an exposed staging server.

It connects with several frontier and locally hosted large language models (LLMs) to automate the process of creating exploits.

The system continuously compares the code of popular web applications before and after each patch announcement.

Once it detects a potential vulnerability, it automatically drafts and executes an exploit without any human involvement.

Exploit features

Framework shows preference for French websites

The exploit_agent framework also comes with scripts for various post-compromise activities, like internal reconnaissance, brute-forcing, and installing or removing webshells.

The operator of this system seems to have a preference for French websites related to media, entertainment, and education.

A comment in the configuration file of the framework, written in Russian, hints at its origin as it asks whoever sets it up to enter keys "right here, not in the chat."

ADVERTISEMENT

Tactical evolution

Fully autonomous attacks on the rise

The GTIG report notes a major shift in AI hacking tactics over the last quarter.

Unlike its previous quarterly tracker, which didn't observe fully autonomous attack pipelines against real targets, this time it has seen such cases.

The researchers also noted that attackers could be more efficient by targeting n-day vulnerabilities (those with existing patches) instead of hunting for zero-days (newly discovered flaws).

ADVERTISEMENT

Exploit generation

PaperCut flaws were exploited as 0-days in late August

The report cites two flaws in print management servers from PaperCut Software Pty Ltd which were exploited as zero-days in late August.

Security researchers and Huntress Labs used large language models (LLMs) to create working exploits within a day of disclosure.

Another campaign tracked by GreyNoise Intelligence four days after disclosure, likely relied on LLMs for its exploits and affected hundreds of organizations.

Framework usage

Open-source agentic penetration testing frameworks used in attacks

The GTIG report also highlights the use of open-source agentic penetration testing frameworks to automate individual steps of live intrusions.

One such incident involved a threat actor using a multi-agent framework called CAIRN against the US aerospace sector.

Another actor ran a customized platform based on PentAGI between late August and early October, scanning various organizations in Asia for vulnerabilities.

ADVERTISEMENT