AI Hacking (Representative image)
Google’s threat researchers say attackers have, for the first time, used an artificial intelligence (AI) framework that works with no human in the loop to turn security patches into working exploits. The finding comes from a new report by Google Threat Intelligence Group (GTIG), titled “Agentic Threat Activity Crosses a Significant Threshold.” The report suggests the gap between a security patch being announced and attackers using it against targets is getting shorter. Patches are meant to protect users, but this new tool reads them to find the flaw they fix.
What is exploit_agent that Google has found
The framework is called exploit_agent. It was found on an exposed staging server, where it was connected to several frontier and locally hosted large language models (LLMs). According to GTIG, its job is to keep comparing the code of popular web applications before and after each patch announcement. When it spots a likely flaw, it drafts an exploit and runs it automatically. The same server also held data taken from victims. It had scripts for brute-forcing, for installing webshells and even for removing webshells left behind by rival attackers. “GTIG identified an exposed staging server hosting a multi-agent vulnerability research and weaponization framework, dubbed exploit_agent, that integrates with multiple frontier and local LLMs. The framework is set up to conduct continuous differential analysis of popular web application software. It compares code before and after patch announcements, analyzes findings to identify potential vulnerabilities, drafts exploits, and automatically executes them without human intervention,” the report explains. “Along with data exfiltrated from targeted victims, we identified multiple scripts capable of facilitating various post-compromise activities, including internal reconnaissance, brute-forcing, installing webshells, and removing other attackers’ webshells,” the report says Victim telemetry suggests the operator was most interested in French websites linked to media, entertainment and education. One comment in the framework’s configuration file, written in Russian, tells whoever sets it up to enter the keys “right here, not in the chat.”
Big shift in just one quarter
In the previous edition of its quarterly tracker, two months ago, GTIG said it had not seen fully autonomous attack pipelines used against real targets. According to the new report, that changed over the past quarter. GTIG also made a point about how attackers may choose their targets. The source material for a patched flaw is limited, so it may be more efficient for attackers to point LLMs at n-day vulnerabilities, which are flaws that already have a patch, than to hunt for zero-days.
How fast can a patch become an attack
The report points to two flaws in print management servers from PaperCut Software Pty Ltd. They were chained together and exploited as zero-days in late August. Security researchers and Huntress Labs Inc. used large language models to produce working exploits within a day of disclosure. Another campaign, tracked by GreyNoise Intelligence Inc. four days after disclosure, is believed to have relied on LLMs to build its exploits. It hit hundreds of organisations. GreyNoise found that one foothold in that campaign reached domain administrator level within minutes. The report describes AI use beyond exploit_agent:Automated intrusion steps: Open-source agentic penetration testing frameworks are being used to automate individual steps of live intrusions. In one intrusion against the US aerospace sector, a threat actor was seen using a multi-agent framework called CAIRN.Wide scanning: Another actor ran a customised platform built on PentAGI between late August and early October. It scanned a wide range of organisations in Asia for vulnerabilities.
Join conversation
Share your thoughts in the comments
Be respectful · TOI community guidelines
Easier to spot, for now: The noisy forensic trail left by unrestricted autonomous execution is said to make these frameworks easier to detect for the time being. GTIG says influence operations now use AI models for far more than writing posts. A turnkey platform that one commercial contractor was building, with help from the Gemini API, uses cloud phone farms and is designed to run more than 100,000 social media accounts.
End of Article
Latest Mobiles
View All
Alcatel V3 Classic
₹12,999
Lava Bold N2 5G
₹12,998
Poco M7 Plus
₹19,999
Moto G06 Power
₹12,360
Realme C83 5G
₹18,999
Itel A90 Limited Edition
Infinix Note 50X
₹13,499
Motorola A200
₹1,149
Itel Zeno 100
₹7,999
OPPO F29 Pro
₹30,999
iQOO Z6 Pro
₹27,890
iQOO Z10R
₹18,772
Poco X7
₹17,999
iQOO Z11 Lite
₹19,999
Lava Shark 2
₹9,150




