Google says an AI framework called exploit_agent automated vulnerability discovery and exploitation with no human in the loop, a first

· CyberSecurity

TLDR

  • Google says a framework called exploit_agent automated vulnerability discovery and exploitation with no human in the loop, a first.1
  • It compared code before and after patches, drafted exploits and ran them, and the server also held stolen victim data.1
  • Google expects unsophisticated attacks needing little human input within three months, and sophisticated autonomous ones within six, BleepingComputer reports.3

Google Threat Intelligence Group (GTIG) said in its fourth-quarter 2026 AI threat tracker, published on 16 November, that in the past quarter it identified threat actors using LLM frameworks "to fully automate vulnerability discovery and exploitation without a human in the loop for the first time".1 GTIG found an exposed staging server hosting a multi-agent vulnerability research and weaponization framework, which it dubbed exploit_agent and which integrates with multiple frontier and local LLMs.1

According to the report, the framework was set up to carry out continuous differential analysis of popular web application software. It compares code before and after patch announcements, analyses the findings to identify potential vulnerabilities, drafts exploits and executes them automatically without human intervention.1 The server also held data exfiltrated from targeted victims and multiple post-compromise scripts for internal reconnaissance, brute-forcing, installing webshells and removing other attackers' webshells.1 The Register, citing the report, says victim telemetry suggested the operator was interested in French websites in the media and entertainment industries and in the education sector.2 GTIG did not name a group or state, CyberScoop reports, and its attribution rests on Russian-language comments in the code, including a configuration file telling the user to enter API keys for "auto-exploitation".4 The Register describes the operator as a suspected Russian-speaking one.2

The report also covers wider use of autonomous multi-agent offensive frameworks, including CAIRN and PentAGI.1 GTIG says it observed CAIRN used in intrusion activity against targets including the US aerospace sector, and tracked a threat actor between late August and early October 2026 using a customised platform built on PentAGI for reconnaissance and vulnerability scanning against a wide range of organisations in Asia.1 It also says a PRC-nexus group was developing its own automated attack framework, likely for espionage.1 For now, GTIG says, these tools trade stealth for velocity, because unrestricted autonomous execution generates forensic traffic and leaves recognisable code-level traces.1

On speed, GTIG says zero-day exploitation rose from an average of 8 a month in 2025 to 11 a month in 2026, before spiking to 22 in August.1 BleepingComputer and CyberScoop both point to PHP2Shell, where hundreds of purported exploits, mostly AI-generated according to BleepingComputer, appeared within days of the 21 October PHP patches, adding verification work for defenders.34 BleepingComputer also reports that GTIG saw UNC6836, which has suspected links to "Midnight Neptune", target cryptocurrency firms in North America and Europe by altering Claude Code settings so the app downloaded the SOMBERMEME backdoor at launch.3

Why it matters: GTIG says LLM-driven pipelines are cutting the time between a patch being published and a weaponised exploit appearing, which leaves defenders less time to apply fixes.1

Sources

  1. GTIG AI Threat Tracker: Agentic Threat Activity Crosses a Significant Threshold (Google Cloud Blog (Google Threat Intelligence Group))
  2. Google: AI agent turned a security patch into a working exploit with no human help (The Register)
  3. Google warns that fully-automated AI attacks are coming soon (BleepingComputer)
  4. Google says Russian-speaking hackers ran the first fully automated AI cyberattack (CyberScoop)