News
General
5 views
Google warns hackers are using AI agents to launch cyberattacks at machine speed
Nov 18, 2026
📍 Phliadelphia,PA, USA
# Google Warns AI-Powered Cyberattack Frameworks Are Helping Hackers Exploit Vulnerabilities Faster
Google’s latest AI threat tracker report has warned that cybercriminals and suspected espionage groups are increasingly using autonomous artificial intelligence systems to identify software vulnerabilities and develop attacks with limited or no human involvement.
According to the report, malicious actors are deploying large language model (LLM)-based frameworks to monitor software updates, examine newly patched security flaws and develop exploits targeting those weaknesses. The approach allows attackers to automate several stages of cyber operations and potentially launch attacks at much greater speed.
Google Threat Intelligence Group (GTIG) said threat actors are moving beyond manually prompting AI tools and increasingly adopting autonomous, multi-agent offensive frameworks. These include ARTEX, CAIRN, CyberStrikeAI and PentAGI, which can support multistage intrusion activities.
The misuse of open-source, agentic penetration-testing frameworks is emerging as a way to automate individual phases and operational tasks during active cyber intrusions, the report noted.
Google described the development as a significant shift toward cyberattacks operating at machine speed. By combining different attack methods and automating tasks, malicious groups can expand the scale of their operations, even though many of the underlying techniques remain familiar.
## AI Agents Expand the Range of Cyber Operations
The report also identified other tools and services being used in malicious activities. These include ZeroHollow, an inference service used for exploit development and LLM routing, and Agent Zero, an open-source agent framework that supports autonomous code execution and dynamic tool generation.
GTIG said financially motivated cybercriminals and suspected espionage groups aligned with China were among the actors using these technologies.
Another emerging application involves automating time-consuming security research through LLM-powered pipelines. These systems can assist with vulnerability analysis and the development of proof-of-concept (PoC) exploits, reducing the time required to turn publicly known security weaknesses into operational attacks.
## Patched Vulnerabilities Remain a Major Target
Google warned that attackers may increasingly use AI tools to weaponize patched vulnerabilities, commonly known as n-day vulnerabilities, before organizations have completed their security updates.
LLMs can help attackers examine a limited collection of technical materials, including different versions of software source code, pre-release or preview builds, security advisories, release notes and publicly available PoC code.
By comparing these resources, malicious actors can identify what a software update changed and investigate whether the underlying weakness can still be exploited in systems that have not been updated.
This approach could shorten the period between the disclosure of a security patch and the exploitation of the vulnerability. Organizations that delay installing updates may therefore face greater exposure as automated tools make vulnerability research faster and more accessible.
## Stolen AI Credentials Create Another Revenue Stream
The report also highlighted a growing threat targeting AI infrastructure itself. Cybercriminals are using a technique known as LLMjacking, which involves stealing access credentials and cloud computing resources from compromised enterprise environments.
The stolen resources can then be resold through underground proxy services, including APIMart and JetNight, allowing other users to access AI models and services at discounted rates.
These underground services are also referred to as LLM mixers. The practice creates an additional source of illicit income while exposing businesses to unauthorized use of their computing resources and potential financial losses.
Google said the growing availability of open-source agentic frameworks has lowered the technical barriers for malicious actors. Such tools make it easier to incorporate new exploits, automate attack processes and develop convincing phishing campaigns.
## Google Calls for Greater Use of Agentic Cybersecurity
John Hultquist, chief analyst at Google Threat Intelligence Group, said the growing use of AI does not necessarily mean attackers are developing entirely new methods of breaking into computer systems.
“They are still going after known, unfixed issues, just with much greater speed and scale,” Hultquist said.
He argued that security teams must increasingly adopt automated, agent-based defensive technologies to keep pace with the speed of modern attacks.
“Instead of relying on manual effort to keep up, we will have to embrace agentic defense. Or we risk falling behind,” he added.
The report highlights an evolving cybersecurity challenge: attackers are using AI primarily to accelerate familiar techniques, automate research and expand the number of operations they can conduct. As these capabilities become more accessible, organizations may need to strengthen vulnerability management, accelerate software patching and adopt automated security monitoring to reduce their exposure.
Google’s latest AI threat tracker report has warned that cybercriminals and suspected espionage groups are increasingly using autonomous artificial intelligence systems to identify software vulnerabilities and develop attacks with limited or no human involvement.
According to the report, malicious actors are deploying large language model (LLM)-based frameworks to monitor software updates, examine newly patched security flaws and develop exploits targeting those weaknesses. The approach allows attackers to automate several stages of cyber operations and potentially launch attacks at much greater speed.
Google Threat Intelligence Group (GTIG) said threat actors are moving beyond manually prompting AI tools and increasingly adopting autonomous, multi-agent offensive frameworks. These include ARTEX, CAIRN, CyberStrikeAI and PentAGI, which can support multistage intrusion activities.
The misuse of open-source, agentic penetration-testing frameworks is emerging as a way to automate individual phases and operational tasks during active cyber intrusions, the report noted.
Google described the development as a significant shift toward cyberattacks operating at machine speed. By combining different attack methods and automating tasks, malicious groups can expand the scale of their operations, even though many of the underlying techniques remain familiar.
## AI Agents Expand the Range of Cyber Operations
The report also identified other tools and services being used in malicious activities. These include ZeroHollow, an inference service used for exploit development and LLM routing, and Agent Zero, an open-source agent framework that supports autonomous code execution and dynamic tool generation.
GTIG said financially motivated cybercriminals and suspected espionage groups aligned with China were among the actors using these technologies.
Another emerging application involves automating time-consuming security research through LLM-powered pipelines. These systems can assist with vulnerability analysis and the development of proof-of-concept (PoC) exploits, reducing the time required to turn publicly known security weaknesses into operational attacks.
## Patched Vulnerabilities Remain a Major Target
Google warned that attackers may increasingly use AI tools to weaponize patched vulnerabilities, commonly known as n-day vulnerabilities, before organizations have completed their security updates.
LLMs can help attackers examine a limited collection of technical materials, including different versions of software source code, pre-release or preview builds, security advisories, release notes and publicly available PoC code.
By comparing these resources, malicious actors can identify what a software update changed and investigate whether the underlying weakness can still be exploited in systems that have not been updated.
This approach could shorten the period between the disclosure of a security patch and the exploitation of the vulnerability. Organizations that delay installing updates may therefore face greater exposure as automated tools make vulnerability research faster and more accessible.
## Stolen AI Credentials Create Another Revenue Stream
The report also highlighted a growing threat targeting AI infrastructure itself. Cybercriminals are using a technique known as LLMjacking, which involves stealing access credentials and cloud computing resources from compromised enterprise environments.
The stolen resources can then be resold through underground proxy services, including APIMart and JetNight, allowing other users to access AI models and services at discounted rates.
These underground services are also referred to as LLM mixers. The practice creates an additional source of illicit income while exposing businesses to unauthorized use of their computing resources and potential financial losses.
Google said the growing availability of open-source agentic frameworks has lowered the technical barriers for malicious actors. Such tools make it easier to incorporate new exploits, automate attack processes and develop convincing phishing campaigns.
## Google Calls for Greater Use of Agentic Cybersecurity
John Hultquist, chief analyst at Google Threat Intelligence Group, said the growing use of AI does not necessarily mean attackers are developing entirely new methods of breaking into computer systems.
“They are still going after known, unfixed issues, just with much greater speed and scale,” Hultquist said.
He argued that security teams must increasingly adopt automated, agent-based defensive technologies to keep pace with the speed of modern attacks.
“Instead of relying on manual effort to keep up, we will have to embrace agentic defense. Or we risk falling behind,” he added.
The report highlights an evolving cybersecurity challenge: attackers are using AI primarily to accelerate familiar techniques, automate research and expand the number of operations they can conduct. As these capabilities become more accessible, organizations may need to strengthen vulnerability management, accelerate software patching and adopt automated security monitoring to reduce their exposure.
Tags
news
Comments (0)
Login to post comments
No comments yet
Be the first to share your thoughts about this post.