Google finds autonomous agents exploiting patched flaws
Google Threat Intelligence Group and Mandiant have identified attackers using AI agents to automate vulnerability discovery and exploitation without human intervention. Researchers found a framework called exploit_agent on an exposed staging server. It compares web application code before and after patches, identifies the underlying flaws, generates exploits and executes them automatically. The server also contained stolen victim data and tools for reconnaissance, brute-forcing and installing webshells. Google did not attribute the activity to a named group or country.
The findings highlight a shrinking window for organizations to address vulnerabilities after fixes become available. Google also observed attackers adopting open source offensive frameworks, although many purported AI-generated exploits do not work. Its claim of newly observed full automation applies specifically to vulnerability discovery and exploitation against web applications, not a major enterprise platform.
Separately, Google described a suspected North Korea-linked cluster, MIDNIGHT NEPTUNE, using AI-generated recruiter identities and fraudulent meeting platforms to deliver ClickFix lures. Attackers modified AI coding assistant settings to download a backdoor at startup. Google says autonomous attack frameworks currently prioritize speed over stealth, leaving detectable traces. Defensive priorities include faster patch deployment, oversight of coding agents and detection focused on attacker behavior rather than framework names.