LIVE

Google Reports First Instance of Fully Autonomous AI Exploitation Tool

🧠 AI Analysis 2 min read 👁 7

Google's threat intelligence teams have uncovered the first known instance of hackers deploying a fully autonomous AI agent to analyze software patches, write exploits, and launch attacks without human intervention.

Google Reports First Instance of Fully Autonomous AI Exploitation Tool🧠AI Analysis

Photo: thenextweb.com

Key points

  • Google discovered an autonomous AI exploit tool named exploit_agent on an exposed staging server.
  • The tool analyzes pre- and post-patch code, writes exploits, and executes them without human intervention.
  • Threat groups are increasingly adopting open-source AI hacking frameworks and stealing cloud credentials for unauthorized model usage.

Security researchers at Google Threat Intelligence Group and Mandiant revealed on Monday that malicious actors have successfully utilized artificial intelligence to automate the entire software exploitation lifecycle. Discovered on an exposed staging server, the tool—dubbed exploit_agent—interfaces with multiple frontier and local AI models to inspect the differences between pre-patch and post-patch web software, identify underlying vulnerabilities, craft functional exploits, and execute them independently.

According to the latest quarterly report, this marks a significant shift from previous threat landscapes where fully automated attack pipelines were theoretical. The compromised server contained stolen victim data, password-guessing scripts, webshell deployment tools, and configuration files featuring comments in Russian. Investigators noted that the operators appeared to focus their campaigns primarily on French media, entertainment, and educational institutions.

Google noted that analyzing recently patched vulnerabilities may offer a streamlined pathway for automated AI systems, as models can easily process fixed codebases, public advisories, and security notes. Beyond autonomous agents, the report detailed a rise in malicious actors hijacking cloud and AI credentials to run unauthorized workloads, with single-day fraudulent charges exceeding $100,000 in some instances. Additionally, researchers observed state-backed and financially motivated groups leveraging open-source AI hacking frameworks like CAIRN against targets including the US aerospace industry.

What to watch next: Monitor whether cybersecurity firms and platform providers can deploy defensive AI countermeasures fast enough to outpace the adoption of autonomous exploit frameworks.

Source thenextweb.com — thenextweb.com/news/google-gtig-ai-exploits-patches-no-human ↗

Contact the newsroom

Got a tip, a proposal or a partnership question? Write to us and we will reply shortly.