Agentic AI Threats Have Crossed a 'Significant Threshold,' Google Warns

https://i.extremetech.com/imagery/content-types/02qm4zbJorkq0xQkKB5sPSA/hero-image.fill.size_1200x675.png

The Google Threat Intelligence Group (GTIG) has discovered an AI tool that can take security updates and turn them into working hacks automatically without any human involvement. The exploit_agent framework ran on an open staging server and used multiple LLMs to check software code before and after it was updated.

In its quarterly cyber threat analysis, the GTIG says the system looked at two versions of code, figured out the weakness the update fixed, created an exploit, and ran it by itself. It's the first time the GTIG has observed threat actors fully automating both vulnerability discovery and exploitation, marking the crossing of a "significant threshold" in agentic AI's impact on cybersecurity, the researchers write.

The framework focused on flaws that vendors had already provided fixes for, allowing attackers to seek out software weaknesses through code changes, release notes, and more.

The GTIG discovered that the attackers had stolen...

Copyright of this story solely belongs to www.extremetech.com. To see the full text click HERE

Read more