Google Discovers Russian-Annotated AI Hacking Tool Capable of Autonomous Operation
Google's Threat Intelligence Group has for the first time discovered threat actors leveraging large language model frameworks to achieve full automation of vulnerability discovery and exploitation without human intervention. The tool monitors software patch updates, compares code differences to pinpoint vulnerabilities, and automatically writes exploit code, while also possessing internal reconnaissance and brute-force capabilities. The report notes a significant increase in zero-day exploitation incidents in 2026, with open-source AI attack tools such as CAIRN and PentAGI also being used in actual intrusions.
Why read this
AI score 68
Google reveals the first fully automated vulnerability exploitation AI tool, exposing how LLMs are being used to automate cyberattacks.
Topics
Body · AI translation
The tool integrates multiple large language models (LLMs), the artificial intelligence technology behind chatbots, including leading commercial models and what Google calls "local" models.
Subsequently, the tool monitors patch updates for popular web software, which are updates released by software developers to fix vulnerabilities.
Once a new patch is released, the tool compares the program code before and after the fix to determine where the vulnerability lies. This is similar to comparing two versions of a document to see what editors have changed, with those modifications revealing the problem.
Afterward, the tool automatically writes exploit code to take advantage of the vulnerability and executes it.
Google stated that this is the first time the company has discovered attackers using this method to carry out attacks.
The report stated: "In the past quarter, the Google Threat Intelligence Group (GTIG) identified for the first time that threat actors leveraged large language model frameworks to achieve fully automated vulnerability discovery and exploitation without human involvement."
Federal Reserve Governor Michael Barr said on Monday that autonomous AI agents could trigger bank runs within minutes.
In September, Torsten Slok, chief economist at Apollo Global Management, warned that banks could lose significant amounts of low-cost deposits if AI assistants automatically transferred household funds to accounts offering higher yields.
An image in Google's report shows the tool examining the access control mechanisms of version 2.19.25 of the WordPress plugin Spectra (Ultimate Addons for Gutenberg).
Access control refers to the code that determines which actions different users can perform. The report did not specify whether this vulnerability check ultimately led to an actual attack.
This pre-production server also stored other scripts capable of mapping internal network structures after attackers breached network defenses, which Google termed "internal reconnaissance."
These scripts were also able to conduct so-called "brute force" attacks by repeatedly attempting numerous passwords and implant Webshells—hidden remote control programs—on websites, allowing attackers to regain access to compromised systems at a later time.
Google also found data stolen from victims on the same server. Related records indicated that the tool's operators may have targeted websites in France's media, entertainment, and education sectors.
API keys are similar to passwords and allow software to access AI services.
Google's translation of the relevant Russian comments read: "LLM API key for automated exploitation (exploit_agent pilot project)," and "Please enter the key yourself—right here, do not type it into the chat." Both lines were marked as required fields.
This finding aligns with broader trends previously described by Google.
Between January and August 2026, the total number of vulnerability disclosures and exploitation incidents nearly doubled.
Zero-day exploits, which are attacks targeting vulnerabilities unknown to software developers, averaged eight occurrences per month in 2025, rose to an average of 11 per month in 2026, and surged to 22 in August.
Google also stated that open-source AI attack tools named CAIRN and PentAGI had been detected being used in actual cyber intrusion activities.
AI Trends