← Voltar ao feed

Google GTIG observes AI agent exploit_agent autonomously diffing patches and building n-day exploits without human input

1 fonte(s) · 1 atualização(ões) · publicado 17/11/2026 11:27 · coletado há 3 horas

Google Threat Intelligence Group (GTIG) reported what it describes as the first fully autonomous attack workflow. A threat actor used an LLM framework called exploit_agent, configured with Russian-language tooling, to compare pre- and post-patch code, identify vulnerabilities and build exploits with no human in the loop. GTIG says actors are increasingly using LLMs and agentic frameworks to automate patch reverse engineering, so they can weaponise n-day flaws quickly and scale exploitation against internet-facing servers.

Financially motivated and likely espionage actors were also seen using open-source agentic pentesting tools (CAIRN, PentAGI, CyberStrikeAI). CAIRN was used in intrusion activity that included a US aerospace target. State-backed groups and "Influence-as-a-Service" vendors use LLMs for pipelines, infrastructure design and detection evasion. In one case, actors used the Gemini API to manage more than 100,000 synthetic accounts through cloud phone farms. China-aligned actors used LLMs to process stolen data, and North Korea-aligned actors targeted cryptocurrency firms with AI-generated personas.

The article is a news summary. No CVEs, IOCs or patches are named, and no web research was performed.

Detalhes técnicos

  • Differential analysis of patches by an LLM agent to produce n-day exploits.
  • Open-source agentic pentesting tools used to automate attack steps.
  • Attempts to manipulate the configuration of autonomous coding assistants.

Linha do tempo

  1. digit.fyi ↗ · pub. 17/11/2026 11:27 · coletado 17/11 13:00 primeiro registro

    Google GTIG observes AI agent exploit_agent autonomously diffing patches and building n-day exploits without human input

    ler a análise desta fonte

    Google Threat Intelligence Group (GTIG) reported what it describes as the first fully autonomous attack workflow. A threat actor used an LLM framework called exploit_agent, configured with Russian-language tooling, to compare pre- and post-patch code, identify vulnerabilities and build exploits with no human in the loop. GTIG says actors are increasingly using LLMs and agentic frameworks to automate patch reverse engineering, so they can weaponise n-day flaws quickly and scale exploitation against internet-facing servers.

    Financially motivated and likely espionage actors were also seen using open-source agentic pentesting tools (CAIRN, PentAGI, CyberStrikeAI). CAIRN was used in intrusion activity that included a US aerospace target. State-backed groups and "Influence-as-a-Service" vendors use LLMs for pipelines, infrastructure design and detection evasion. In one case, actors used the Gemini API to manage more than 100,000 synthetic accounts through cloud phone farms. China-aligned actors used LLMs to process stolen data, and North Korea-aligned actors targeted cryptocurrency firms with AI-generated personas.

    The article is a news summary. No CVEs, IOCs or patches are named, and no web research was performed.

    Technical Details

    • Differential analysis of patches by an LLM agent to produce n-day exploits.
    • Open-source agentic pentesting tools used to automate attack steps.
    • Attempts to manipulate the configuration of autonomous coding assistants.