← Voltar ao feed

GTIG: Threat Actors Use exploit_agent and Other AI Frameworks to Autonomously Discover and Exploit Vulnerabilities

1 fonte(s) · 1 atualização(ões) · publicado 17/11/2026 15:16 · coletado há 49 minutos

Google Threat Intelligence Group (GTIG) reports that state-sponsored and cybercrime actors are increasingly adopting generative AI and multi-agent offensive frameworks to speed up and scale intrusions. For the first time, GTIG identified actors using an AI framework, exploit_agent, to automate both vulnerability discovery and exploitation without human intervention. It was found on an exposed staging server. The framework compares patched and unpatched web application code, identifies flaws, develops exploits and runs them automatically. Evidence pointed to interest in French media and entertainment websites and the education sector.

Detalhes técnicos

  • The staging server also held stolen victim data and scripts for post-compromise activity, including reconnaissance and webshell deployment.
  • Frameworks in use include Artex (the most used, mainly by Chinese-speaking groups), Cairn, PentAGI, CyberStrikeAI and HexStrike AI. They are used to reverse engineer patches and weaponize n-day vulnerabilities at machine speed.
  • Cairn was seen in activity against the US aerospace sector. PentAGI was used for reconnaissance and vulnerability scanning across organizations in Asia.
  • A Chinese group was building its own automated attack framework, likely for espionage.
  • Other tools and services observed: ZeroHollow (inference service for exploit development and LLM routing) and Agent Zero.
  • State actors using LLMs include Leaky Castle and other Chinese APTs, APT42 (Calanque Ion) and Midnight Neptune (UNC1069).
  • The Jieyou PhaaS has been active since March 2026 and offers a chatbot for configuring phishing panels.
  • LLMJacking and LLM Mixer marketplaces resell stolen LLM access. Cumulative abuse reached $1M in one case.
  • Influence-operation vendors are using LLMs for content generation and automation.

Remediação

The article gives no specific remediation. A quoted commentator advises getting the cybersecurity fundamentals right, including faster patching, and moving toward automated detection and response.

Linha do tempo

  1. securityweek.com ↗ · pub. 17/11/2026 15:16 · coletado 17/11 16:00 primeiro registro

    GTIG: Threat Actors Use exploit_agent and Other AI Frameworks to Autonomously Discover and Exploit Vulnerabilities

    ler a análise desta fonte

    Google Threat Intelligence Group (GTIG) reports that state-sponsored and cybercrime actors are increasingly adopting generative AI and multi-agent offensive frameworks to speed up and scale intrusions. For the first time, GTIG identified actors using an AI framework, exploit_agent, to automate both vulnerability discovery and exploitation without human intervention. It was found on an exposed staging server. The framework compares patched and unpatched web application code, identifies flaws, develops exploits and runs them automatically. Evidence pointed to interest in French media and entertainment websites and the education sector.

    Technical Details

    • The staging server also held stolen victim data and scripts for post-compromise activity, including reconnaissance and webshell deployment.
    • Frameworks in use include Artex (the most used, mainly by Chinese-speaking groups), Cairn, PentAGI, CyberStrikeAI and HexStrike AI. They are used to reverse engineer patches and weaponize n-day vulnerabilities at machine speed.
    • Cairn was seen in activity against the US aerospace sector. PentAGI was used for reconnaissance and vulnerability scanning across organizations in Asia.
    • A Chinese group was building its own automated attack framework, likely for espionage.
    • Other tools and services observed: ZeroHollow (inference service for exploit development and LLM routing) and Agent Zero.
    • State actors using LLMs include Leaky Castle and other Chinese APTs, APT42 (Calanque Ion) and Midnight Neptune (UNC1069).
    • The Jieyou PhaaS has been active since March 2026 and offers a chatbot for configuring phishing panels.
    • LLMJacking and LLM Mixer marketplaces resell stolen LLM access. Cumulative abuse reached $1M in one case.
    • Influence-operation vendors are using LLMs for content generation and automation.

    Remediation

    The article gives no specific remediation. A quoted commentator advises getting the cybersecurity fundamentals right, including faster patching, and moving toward automated detection and response.