[Photo: Shutterstock]

Google has confirmed for the first time a case in which attackers fully automated vulnerability scanning and attacks using an LLM framework without human involvement, SiliconANGLE reported on Nov. 16 local time.

The report said Google Threat Intelligence Group (GTIG) disclosed a framework called exploit_agent in its quarterly report.

The framework was found on a test server that attackers left open to the internet without security settings, and it connects to multiple large language models (LLMs). It repeatedly compared web application code before and after patches, and if it found a vulnerability, it wrote attack code and executed it automatically.

Attackers used the system to target French media, entertainment and education websites.

Victim data and web shell installation scripts were also found on the test server.

GTIG said it had not confirmed a fully autonomous attack in the real world as recently as two months ago. But the situation has changed. GTIG said attackers may have judged it more efficient to use LLMs to exploit already disclosed vulnerabilities rather than finding new vulnerabilities, or zero-days.

The time from vulnerability disclosure to completion of attack code has also fallen sharply.

Google said state-backed groups, private companies and political networks routinely use AI for public opinion manipulation. It said LLMjacking is also widespread, becoming a new revenue source for attackers, while victim organisations are paying large costs.

GTIG predicted that within the next three months, simple attacks carried out by multiple AI agents with almost no human involvement would emerge. It forecast that within six months, sophisticated fully autonomous attacks are also likely to appear.

Keyword

#Google #Google Threat Intelligence Group #exploit_agent #SiliconANGLE #LLMjacking
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.