Google: “AI Develops Attack Code Without Human Involvement” ... Hacking Gets Faster
- Input
- 2026-11-17 09:46:47
- Updated
- 2026-11-17 09:46:47

According to the latest “AI Threat Tracking Report,” released on the 17th by Google Threat Intelligence Group (GTIG), cyberattackers are using autonomous AI agents to automate tasks previously performed by people, such as finding vulnerabilities and developing attack code.
In particular, GTIG said it had identified for the first time an AI framework that analyzes software vulnerabilities and develops attack code without human intervention.
The newly discovered “exploit_agent” is an automation framework that uses multiple large language models (LLMs). It is designed to compare software code from before and after a security patch is applied, identify vulnerabilities, and write and execute exploit code.
Previously, hackers had to analyze security patches to find vulnerabilities and write attack code themselves. But with AI agents automating this process, the time needed to prepare an attack can now be significantly reduced.
Google warned that “N-day” vulnerabilities—those for which security patches have already been made public but have not yet been applied by companies or users—could become major attack targets. This is because AI can analyze code differences before and after a patch to find the vulnerabilities that were fixed.
There are also signs that attackers are using multiple AI agents together. GTIG said autonomous attack frameworks such as ARTEX, CAIRN, CyberStrikeAI, and PentAGI are being used in actual cyberthreat activity.
These tools automate tasks such as finding vulnerabilities, generating attack code, and analyzing attack results, helping attackers carry out multi-stage intrusions more quickly.
A case in which CAIRN was used was identified in an intrusion operation targeting the U.S. aerospace industry.
The use of AI by state-linked hacking groups is also becoming more sophisticated. According to Google, China-linked cyber espionage groups used LLMs to research targets and analyze and process stolen data to produce intelligence reports.
North Korea-linked attackers targeted cryptocurrency-related companies using AI-generated fake personas and recruiter profiles. SynthID watermarks, from Google’s technology for identifying AI-generated content, were also found in some fake profile images.
Supply chain attacks exploiting AI coding tools have also emerged. Attackers manipulated the settings of autonomous coding tools so that malware would be installed when users ran them.
The misuse of AI is expanding beyond cyber intrusions into areas such as public opinion manipulation and the monetization of crime.
GTIG said it had identified cases in which attackers used AI to generate fake accounts and synthetic content at scale and automate the spread of information on online platforms.
“LLM jacking”—the theft of companies’ cloud resources and access to AI services—has also been identified as a new threat. Attackers use stolen accounts to access AI models without authorization or sell that access to other criminals.
[email protected] Jang Min-kwon Reporter