Google finds AI agents autonomously turning security patches into exploits

Covered by 5 sources
Read onThe Hacker News· 3 min
Google's Threat Intelligence Group uncovered attackers using multi-agent AI frameworks like exploit_agent to automatically analyze software patches and generate exploits without human help, targeting sectors including French media and U.S. aerospace. These AI-driven methods accelerate cyberattacks by rapidly weaponizing vulnerabilities and have been linked to groups including Chinese threat actors.

What happened

Google's Threat Intelligence Group has discovered that hackers are using autonomous AI frameworks, such as exploit_agent, to automatically analyze software patches and create exploits without human involvement. These AI-driven tools have been used to target French media, education, and entertainment websites, as well as sectors like U.S. aerospace. The attacks involve multi-stage operations, including credential theft and the use of compromised cloud resources. The findings represent the first observed use of fully autonomous attack pipelines in real-world cyberattacks.

Key details

  • The exploit_agent framework uses multiple large language models to analyze code changes from software patches and generate exploits autonomously.
  • Victims include French media, education, and entertainment websites, and U.S. aerospace organizations.
  • Open-source agentic frameworks like ARTEX, CAIRN, CyberStrikeAI, and PentAGI are being used to automate different phases of attacks.
  • Attackers have stolen AI and cloud credentials, with some fraudulent charges exceeding $100,000 in a single day and up to $1 million cumulatively.
  • Illicit marketplaces have emerged, selling unauthorized access to AI services for prices ranging from $0.10 to $25 per million tokens.

Why it matters

Organizations in media, education, aerospace, and other sectors now face cyberattacks that can be developed and launched much faster, reducing the time defenders have to respond to new vulnerabilities. Companies using cloud and AI services are at risk of financial losses from stolen credentials and unauthorized resource use. Security teams must now verify more exploit claims and contend with increased attack speed and scale, while also having new forensic traces to help with detection.

Also covered by 4 sources


More in Cybersecurity