Advertisement

Google Warns 'Era Of Autonomous AI Attacks Has Arrived'

Google's Threat Intelligence Group says it discovered an AI-powered framework that can analyse software patches, identify vulnerabilities, write exploits and execute them without human intervention.

Google Warns 'Era Of Autonomous AI Attacks Has Arrived'
AI-generated image

Google has warned that, "the era of autonomous attacks has arrived."

Google's Threat Intelligence Group (GTIG) says the era of autonomous cyberattacks has arrived, with malicious actors beginning to use AI-powered systems that can identify software vulnerabilities, develop exploits and execute attacks without human intervention.

In its latest AI Threat Tracker report, the company detailed an AI-powered framework called exploit_agent, which it found on an exposed staging server.
"In the past quarter, GTIG identified threat actors using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop for the first time," Google said.

How Does This AI Hacker Actually Work?

The framework, which Google calls exploit_agent, was discovered on an exposed staging server and was designed to automate several stages of a cyberattack.

It uses multiple large language models (LLMs), including frontier and locally hosted models, to continuously examine popular web applications for vulnerabilities.

When a software company releases a security update, the framework compares the code before and after the update. By examining what has changed, it attempts to identify the vulnerability that the developer has fixed. It then drafts an exploit designed to take advantage of that weakness and automatically executes it.

Basically, the AI system is attempting to reverse-engineer a security fix and turn it into an attack.

"It compares code before and after patch announcements, analyzes findings to identify potential vulnerabilities, drafts exploits, and automatically executes them without human intervention," Google describes.

When a company discovers a security vulnerability, it typically releases a software patch to fix it. However, the release of that patch can also give hackers clues about the original weakness.

By comparing the vulnerable version with the corrected version, attackers can work backwards to understand what was fixed and attempt to exploit systems that have not yet been updated.

These are known as n-day vulnerabilities - flaws that have already been disclosed or patched, but may remain unpatched on some systems.
To be sure, this is different from a zero-day attack, which exploits a vulnerability before a fix is available.

This distinction is important here because exploit_agent is not necessarily discovering previously unknown weaknesses. It is automating the process of finding and exploiting flaws that have already been fixed in software updates.

What Did Google Actually Find On The Server?

The discovery was not limited to the AI framework itself. Google said the exposed server also contained data exfiltrated from targeted victims, along with scripts designed to facilitate activities after an initial compromise.

These included internal reconnaissance, brute-forcing, installing webshells and even removing webshells left behind by other attackers. A webshell is a malicious script that gives an attacker a way to remotely control or interact with a compromised web server.

Google's analysis of victim telemetry suggested that the operator was particularly interested in French websites in the media and entertainment industries, as well as the education sector.

"Along with data exfiltrated from targeted victims, we identified multiple scripts capable of facilitating various post-compromise activities...," Google said.
So, Google found evidence of compromised victim data and post-compromise tooling, not merely a theoretical AI hacking experiment.

However, Google does not disclose the number of affected organisations, identify the victims or name the operator behind exploit_agent. It also does not explicitly establish that a particular exploit generated by this framework was responsible for the stolen data. 

AI-Powered Hacking Was Already Getting Faster

Google's latest discovery follows a series of incidents in which hackers have increasingly used AI to automate different stages of cyberattacks.

In September 2026, cybersecurity firm GreyNoise reported that a threat actor had deployed hundreds of AI agents in a campaign targeting vulnerable PaperCut print-management servers.

The operation used OpenAI's Codex harness alongside a DeepSeek model and other offensive security tools to develop, test and deploy exploits against vulnerable systems.

According to GreyNoise, the campaign compromised at least 440 PaperCut instances belonging to 395 organisations across 48 countries. The attackers harvested credentials from 280 victims and obtained operating system or domain secrets from 147.

"The adversary went from an empty workspace to first achieving RCE against a real victim in just under four hours," GreyNoise noted. RCE stands for remote code execution, which means an attacker can run malicious code on a target system remotely.

GreyNoise also reported that the attacker compromised at least 11 organisations in just 26 seconds once the wider campaign began.

The PaperCut operation involved hundreds of AI agents and automated workflows, whereas Google's exploit_agent discovery is specifically significant because the framework was designed to automate vulnerability discovery and exploitation without a human in the loop.

Google's earlier September 2026 threat research had already documented an AI-assisted credential-harvesting operation in which attackers planned, built and executed a campaign in under six hours after compromising a cloud resource. The operation used an AI coding chatbot and preconfigured agent instructions to automate scanning and credential harvesting.

The chronology is hair-raising - AI-assisted attacks, followed by increasingly autonomous workflows, and now Google's first documented instance of a fully automated vulnerability-discovery and exploitation pipeline.

But Are AI Hackers Already Unstoppable?

Despite the alarming discovery, Google acknowledges that autonomous AI hacking frameworks still have limitations.

The company says these tools currently trade stealth for speed. Their unrestricted automated execution generates forensic traffic, recognisable identifiers and distinct code-level traces, making them easier for security teams to detect and potentially disrupt.

Google also notes that the growing use of AI has created a flood of exploit claims that do not actually work, adding to the verification burden on cybersecurity professionals.

"Notably, while these frameworks are capable of significantly compressing attack timelines to machine speed and achieving operational objectives, they currently trade stealth for velocity," Google said.

According to Google's findings, between January and August 2026, overall vulnerability disclosures and exploits nearly doubled. However, zero-day exploits rose from a monthly average of eight in 2025 to 11 in 2026, before reaching 22 in August.

Google Says Defenders Will Need AI Too

Google believes cybersecurity teams will increasingly need to deploy AI agents of their own to keep pace with automated attacks.

John Hultquist, Chief Analyst at Google Threat Intelligence Group, argued that attackers are still exploiting known vulnerabilities rather than relying on entirely new techniques. The difference is the speed and scale at which they can now operate.

"Instead of relying on manual effort to keep up, we will have to embrace agentic defense. Or we risk falling behind," Hultquist said.
Google is also promoting its AI Threat Defense offering, which combines Gemini, Wiz, CodeMender and Mandiant capabilities to help identify and fix security weaknesses.

It is worth noting that Google has a commercial interest in the cybersecurity market, and its announcement promotes its own defensive products alongside its threat findings. That, of course, doesn't invalidate the research, but it is an important disclaimer.

Show full article

Track Latest News Live on NDTV.com and get news updates from India and around the world

Follow us:
Listen to the latest songs, only on JioSaavn.com