
What happened
Google's Threat Intelligence Group found an exposed server running "exploit_agent," an AI framework that connects to multiple language models, compares code before and after patches, and builds and runs exploits automatically.
Why it matters
This is the first time vulnerability discovery and exploitation have been fully automated with no human in the loop, which the group calls a critical turning point — though the attacks now trade stealth for speed and leave forensic traces.
WHO IT HITSSecurity teams defending web-facing software now face attacks that can find and weaponize patches at machine speed, potentially without any human operator — though the current automated attacks leave forensic traces that defenders may be able to use.
Summaries like this, every morning.
The finding marks a shift from AI-assisted hacking to fully automated exploitation. Google's Threat Intelligence Group identified an exposed server whose configuration points to an AI framework, "exploit_agent," that ties into multiple language models and operates without a human in the loop — comparing pre- and post-patch code to find and weaponize vulnerabilities.
The server data suggests the operator focused on French websites in media, entertainment, and education, and Russian-language comments in the configuration files indicate the operator speaks Russian. Google describes the development as a critical turning point, while noting that today's automated attacks trade stealth for speed and leave forensic traces that make them easier to detect.
Google places current activity at Level 5, where AI tools carry out several stages of an attack with minimal human involvement, and says more powerful attacks are only months away — a signal that the window before fully autonomous attacks become both fast and hard to spot may be short.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Nikkei xTECH published a beginner-oriented, step-by-step tutorial on implementing an approval-request function…

Felix Rieseberg, who works on Claude at Anthropic, said on X that Claude can now build "bizarre & beautiful si…

Moonshot AI is redirecting its Kimi language models from coding toward biology, chemistry, and pharma research…

Apple's iOS 27.3, iPadOS 27.3, and macOS 27.3 betas test daily usage limits for four categories of Siri AI and…

At a Monday briefing in Washington, DC, OpenAI announced a classifier it says better detects abusive imagery o…

Harvard-led researchers used satellite imagery to identify 403 US hyperscale data centers and estimated they b…
