Security

AI Frameworks Now Autonomously Exploit Vulnerabilities in Live Attacks

Google researchers document threat actors using multi-agent systems that discover, weaponize, and deploy exploits without human oversight.

Omega Editorial · November 17, 2026 · 3 min read

Key takeaways

  • Google documented the first autonomous AI framework (exploit_agent) that discovers vulnerabilities and executes exploits without human intervention in live attacks.
  • Multiple threat actor groups, including Chinese APTs and Iranian state-sponsored hackers, have adopted AI frameworks like Artex, Cairn, and PentAGI for offensive operations.
  • Attackers are using LLMs to reverse engineer patches at machine speed and rapidly weaponize n-day vulnerabilities, fundamentally changing the speed of intrusions.
  • LLMJacking operations have generated up to $1 million in resource abuse in single cases, spawning illicit marketplaces for stolen LLM access.
  • Organizations with monthly patching cycles face a critical mismatch against adversaries now operating at machine speed.

Threat actors have crossed a significant threshold in offensive automation: Google's Threat Intelligence Group has documented the first confirmed use of an AI framework that autonomously discovers vulnerabilities and exploits them in live attacks without human intervention.

Researchers discovered a multi-agent system called exploit_agent on an exposed staging server. The framework was configured to compare patched and unpatched web application code, identify security flaws, develop working exploits, and execute them automatically. The same server contained stolen victim data and post-compromise scripts for reconnaissance and webshell deployment, with evidence suggesting targeting of French media, entertainment, and education sectors.

Why it matters

This represents a fundamental shift in the cyber threat landscape. Organizations accustomed to patching on monthly cycles now face adversaries operating at machine speed. The economic equation has changed: sophisticated attack capabilities that once required well-resourced teams and specialized expertise can now be delegated to autonomous agents, placing advanced offensive tools within reach of less sophisticated threat groups. Critical infrastructure operators face particular risk, as their often-limited security resources are increasingly mismatched against automated adversaries.

Widespread Framework Adoption

Google and Mandiant researchers have observed threat actors deploying multiple AI-powered offensive frameworks in active operations. Artex has emerged as the most widely adopted tool, particularly among Chinese-speaking groups. Other frameworks seeing operational use include Cairn, PentAGI, CyberStrikeAI, and HexStrike AI.

Cairn appeared in attacks targeting the U.S. aerospace sector, while PentAGI was used for systematic reconnaissance and vulnerability scanning across Asian organizations. Google also identified a Chinese threat group building a custom automated attack framework, likely for espionage operations.

These frameworks leverage large language models to reverse engineer security patches at machine speed and rapidly weaponize n-day vulnerabilities—flaws for which patches exist but may not yet be deployed.

State-Sponsored and Criminal Adoption

Multiple nation-state groups have integrated LLMs into their operations. Chinese APTs including Leaky Castle, Iran-linked APT42 (Calanque Ion), and North Korean groups such as Midnight Neptune have adopted the technology. The Jieyou phishing-as-a-service operation, active since March 2024, features a model-agnostic chatbot that enables clients to configure phishing panels and modify templates.

Beyond offensive operations, threat actors are stealing LLM credentials through "LLMJacking" campaigns, with one case generating $1 million in cumulative resource abuse. This has spawned LLM Mixers—illicit marketplaces selling access to LLM services through pools of stolen credentials.

The Defense Challenge

Google's Threat Intelligence Group assesses that the widespread LLM adoption introduces forensic noise that can degrade operational security, and may reduce reliability for stealth-focused operators. However, cybercrime actors less concerned with stealth will likely continue incorporating autonomous tools, as increased campaign reach compensates for any decrease in individual success rates.

Matt Hartman, chief strategy officer at Merlin, emphasized the urgency: "Defenders must do two things at once: nail the cybersecurity fundamentals today and accelerate the shift toward automated defenses that can detect and respond at machine speed."

These findings were first reported by Google Threat Intelligence Group in their latest AI threat tracker report, published through SecurityWeek.

  • #ai security
  • #autonomous exploits
  • #threat intelligence
  • #vulnerability exploitation
  • #offensive ai
  • #apt groups

This is an original analysis by the Omega editorial team. Source reporting: AI Watch.

Want systems like this working for your business?

Book a Call

More in Security