Google researchers have for the first time detected an artificial intelligence-based hacker system capable of independently finding software vulnerabilities, creating tools to exploit them, and conducting attacks without human intervention at these stages. The system's technical files were found to contain Russian-language comments, and among its targets were French websites.
This is reported by Politico, citing a Google Threat Intelligence Group study published on November 16.
The specialists discovered an internet-accessible server of the attackers that ran a system called exploit_agent. It combined several AI agents that used different large language models to automate cyberattacks.
A key feature of the system is that it can use software security updates to find ways to hack. The artificial intelligence compares the code before and after a patch, identifies the vulnerability eliminated by developers, creates an exploit for it, and automatically runs it.
Previously, such operations required the direct involvement of skilled hackers who individually analyzed code and adjusted attack tools. Now a significant part of this work can be performed by autonomous AI systems.
On the discovered server, researchers also found data stolen from affected organizations and tools for further penetration into compromised systems. For example, they could be used to search for information in internal networks, guess passwords, and install remote access tools.
According to Google's conclusions, the system's operators showed particular interest in French websites in the media, entertainment, and education sectors. However, the company did not report the number of affected organizations, the volume of stolen information, or the extent of the damage.
The configuration files of exploit_agent contained instructions in Russian. This served as the basis for assuming the Russian-speaking origin of the operators. However, Google did not determine their affiliation with a specific hacking group or state.
The Threat Intelligence Group report notes that the use of autonomous AI agents changes the nature of cyber threats. Such systems allow attackers to reduce the time between the publication of vulnerability information and its use in an attack, while increasing the number of potential targets.
Analysts expect further proliferation of such tools. This creates additional risks for companies using web applications and internet services, as the time to install security updates may be significantly reduced.
Based on materials from: Politico, Google Threat Intelligence Group