siliconangle.com web signal

Google flags 'exploit_agent' as first autonomous AI attack in wild

TL;DR

  • Google TIG says exploit_agent is the first fully autonomous AI attack pipeline it has observed in the wild.
  • The framework diffed web app code before and after patches, drafted exploits using connected frontier and local LLMs, and executed them without human oversight.
  • Targets skewed to French media, entertainment and education sites; configuration files carried Russian-language comments.

A framework called exploit_agent sat on an exposed staging server, wired into several frontier and locally hosted large language models, continuously diffing popular web application code before and after each patch and running the exploits it drafted with no human in the loop. SiliconAngle summarizes the Google Threat Intelligence Group finding bluntly: "Once a likely flaw is identified, the exploit it drafts is executed automatically."

That it exists at all is what GTIG is calling a first. Two months ago, the group said it had not seen fully autonomous attack pipelines deployed in the wild. This quarter's report, titled "Agentic Threat Activity Crosses a Significant Threshold," says that changed.

The operator left a mess. Data taken from victims sat on the same server as scripts for brute-forcing, installing webshells "and even removing webshells left by rival attackers." One comment in the framework's configuration file, written in Russian, tells whoever set it up to enter the keys "right here, not in the chat." "Victim telemetry suggests the operator was most interested in French websites tied to media, entertainment and education."

GTIG offers a theory for why autonomous frameworks may lean this way: "attackers may find it more efficient to point LLMs at n-day vulnerabilities than to hunt for zero-days." The same report tracks open-source agent frameworks surfacing in other intrusions: CAIRN in a U.S. aerospace incident, and a customized PentAGI deployment scanning organizations in Asia.

Speed is on the attackers' side. Researchers and Huntress Labs used LLMs to generate working exploits for PaperCut vulnerabilities within a day of disclosure. GreyNoise Intelligence tracked a campaign four days after disclosure that hit hundreds of organizations, with one foothold reaching domain administrator within minutes. The partial comfort for defenders: "the noisy forensic trail left by unrestricted autonomous execution is said to make the frameworks easier to detect."

The report lands the same day Anthropic cut its coordinated vulnerability disclosure window from 90 to 30 days for high-severity bugs, one of 199 cybersecurity stories on our tracker in the last 90 days.