According to Google, the phase of autonomous cyberattacks has already begun. In a recent security report, the company stated that criminals are employing artificial intelligence agents capable of locating software flaws, creating malicious code, and launching attacks without needing direct human intervention.
This warning was published by the Google Threat Intelligence Group (GTIG) on Monday, November 16. The group observed a progression in the use of AI by hackers, who have evolved from simple chatbot queries to systems capable of managing various phases of an intrusion.
For Google, the speed of these new tools poses a major challenge to security teams, as they still rely on manually performed analysis and response processes. One notable discovery was a system called exploit_agent, which researchers found on a publicly accessible test server on the internet.
This tool used multiple AI models to examine software updates and detect potential weaknesses. In practice, the system compared old and new versions of the software to discover what problems had been fixed. Based on this information, it generated code to exploit these gaps and conducted tests automatically, without requesting human approval at each step.
Google also recovered data obtained from victims and utilities that could be used after the invasion occurred. Indications suggest a focus on French portals belonging to the media, entertainment, and education segments. The report also mentions the CAIRN and PentAGI tools, which orchestrate different AI agents. According to Google, these tools have already been seen in invasion and vulnerability hunting activities in both the United States and Asia.
This advancement marks a change compared to the AI malware Google identified previously. At that time, experts highlighted the limitations of early attempts to use generative models to modify malicious code. Currently, the main concern is the possibility of agents reducing the time between a vulnerability's disclosure and its exploitation by criminals.
To combat these new threats, Google is expanding the use of automated security resources. The goal is to identify risks and expedite the fixing of flaws. The intention is to enable defense systems to analyze programs, prioritize vulnerabilities, and respond to incidents with a speed close to that of automated attacks.
The subject had been addressed in past investigations. In May, Google announced it had blocked an attack attempt developed with AI assistance. Other corporations have also reported the use of this technology in real attacks. Anthropic reported a campaign involving Claude, where criminals allegedly automated a large part of the operations targeting organizations in various countries.
Despite the progress, the Google report emphasizes that autonomous tools still have limitations, potentially generating ineffective code, making errors, and leaving traces that facilitate the detection of intrusions. Nevertheless, the ability to execute more attacks in less time underscores the extreme need to fix vulnerabilities quickly.

