Security

AI Cyberattacks Enter 'Autonomy Phase,' Forcing Businesses to Rethink Security Response

TECHWORLD ·

AI Cyberattacks Enter 'Autonomy Phase,' Forcing Businesses to Rethink Security Response
[Photo: Google Cloud]

✦ AI Summary

Google Threat Intelligence Group (GTIG) on the 17th released its 'AI Threat Tracker' report. The report said AI use has expanded beyond human assistance to autonomous multi-agent frameworks and automated vulnerability exploitation. GTIG warned of cases of vulnerability exploitation without human intervention, along with shorter attack times and larger scale in the future.

BizCrushWant to follow Korean talks and live streams in real time?Try CapCrush on the web

As cyberattacks using AI continue to grow more sophisticated, attack methods are moving beyond human-assisted operations and into an autonomy phase. Recent industry reports indicate that they have reached a level where vulnerabilities can be detected and attack code can be developed without human intervention, and there have also been ongoing cases in which AI agent-based systems autonomously carry out attack stages. As a result, attack speed and scale are expected to expand with automation.

Google Threat Intelligence Group (GTIG) on the 17th released its 'AI Threat Tracker' report. The report includes analysis of AI use cases observed since its release last September, and the subjects of analysis are cybercrime, espionage activities, and information operations.

According to the report, recent threat actors have gone beyond simple prompt input and are using autonomous multi-agent frameworks, while automated vulnerability exploitation techniques are also being used in attacks. As a result, multiple attack stages that were previously carried out by humans are being automated, and attacks against multiple targets are moving quickly.

Examples of offensive multi-agent frameworks include ARTEX, CAIRN, CyberStrikeAI, and PentAGI. The report also cited a case in which one threat actor used CAIRN during an intrusion targeting the U.S. aerospace industry.

Another focus is shifting to the automation of vulnerability research and exploitation. Threat actors are using LLMs and agentic frameworks to perform differential analysis of code before and after security patches, with the aim of identifying modified vulnerabilities. In this process, reverse engineering is also being fully automated.

Vulnerability discovery through code comparison before and after patches could lead to rapid weaponization of N-day vulnerabilities. It could also enable large-scale automated attacks against internet-exposed servers without human intervention.

GTIG said it had, for the first time, identified a vulnerability exploitation case that was entirely human-out-of-the-loop. GTIG also announced the existence of an LLM framework called 'exploit_agent,' made up of Russian-language tools, and said the framework is designed to detect vulnerabilities through continuous comparison of code before and after patching and to automatically develop attack code without human intervention.

AI-based attacks are also expanding into financially motivated cybercrime. There is a pattern in which LLM credentials are stolen from compromised corporate environments, cloud resources within those environments are also stolen, and the stolen resources are then used to resell access to frontier AI models through the underground proxy service 'LLM mixer.'

The use of LLMs is also increasing in the field of information operations (IO). State-backed threat groups, Influence-as-a-Service providers, and networks of fake accounts that monetize political content are routinely using LLMs, with the goal of mass-producing synthetic media, generating localized personas, and automating content distribution across multiple platforms.

AI use by state-linked hacking groups is also becoming more sophisticated. GTIG said China-linked cyber espionage groups are using LLMs to investigate targets, automate data analysis, and develop new supply chain attack paths. One example presented was a method of manipulating the configuration of an autonomous coding assistant to insert a backdoor when code is executed in a local environment. GTIG said the scope of LLM use by China-linked cyber espionage groups has expanded to target investigation, data analysis automation, and the development of supply chain attack paths.

The report pointed to signs that North Korea-linked attackers used AI-generated fake personas and related tools in real-world attacks. The target was a cryptocurrency company, and some fake persona images contained Google's AI watermark SynthID. Attempts to manipulate the configuration of an autonomous coding assistant were also detected.

Based on this, GTIG emphasized that the adversarial AI threat landscape has entered a major turning point. Examples of deployment include autonomous multi-agent frameworks, automated vulnerability exploitation pipelines, and new monetization methods such as LLMJacking.

GTIG projected that the use of AI agents and automation frameworks in cyberattacks will expand in the future. It also warned that the use of generative AI could shorten attack times, increase operational scale, and enable multi-stage intrusion attacks at machine speed.

Industry officials said that while warning about the potential for AI-driven vulnerability exploitation and the progression toward automated attack execution, security response methods also need to become automated. They said the continuing sophistication of cyberattacks is exposing the limits of existing response methods, and that the importance of securing enterprises' vulnerability management capabilities and their threat detection and response automation capabilities will continue to grow.

Source: TECHWORLD · Kim Hye-jin
Original: https://www.epnc.co.kr/news/articleView.html?idxno=409907

References

This article was produced with the help of an automated content generation algorithm.


Source: TECHWORLD

View original

This article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.

Keep reading

BizCrush
BizCrush showing live captions over an online meeting

BizCrush app

Live captions and summaries for meetings and calls

BizCrush Inc. · CEO Taemin Kwak

HQ 100 Middlefield Rd, Menlo Park, CA 94025, USA

Korea office Z112, Room 602, 6F, 602 Yeongdong-daero, Gangnam-gu, Seoul, Republic of Korea

Support help@bizcrush.ai · Partnerships business@bizcrush.ai

AI Notice

Some features use generative AI.

Those features carry their own pre-use notice and mark AI-generated results. AI output is for reference only — always check the original source and consult an expert before making an important decision.