November 18, 2026, 1:14 pm | Read time: 3 minutes
Cybercriminals can now use AI to largely execute attacks on IT systems autonomously. Google experts have discovered a tool for the first time that automatically searches for security vulnerabilities, creates appropriate attack code, and deploys it without human control. This development could significantly shorten the time between discovering a vulnerability and launching an attack.
AI autonomously searches for security vulnerabilities
The Google Threat Intelligence Group (GTIG) publicly shared details about an AI framework called “exploit_agent” on November 16, 2026. Google discovered the tool on an unprotected server, which also contained stolen data and scripts for further attacks. According to the experts, it is the first observed framework that automates an attack so comprehensively without human intervention.
To do this, the system uses several AI models. It compares the program code of common web software before and after a security update. This way, it aims to identify potential vulnerabilities. If the framework finds a gap, it independently creates the necessary attack code and executes it.
This allows the software to continuously search for new attack targets. Merritt Baer, former deputy chief security officer at Amazon Web Services, describes the difference this way: “Attackers haven’t invented anything new, they’ve just automated the tedious part. What used to take days, an agent now does overnight.”
More AI tools are already in use
In addition to “exploit_agent,” Google has observed other AI frameworks like CAIRN in actual attacks. Targets included sectors such as the U.S. aerospace industry. Google suspects both financial and state motives behind the attacks.
Cybercriminals are also targeting access data for cloud and AI services. With stolen data, they can use the models at the expense of their victims. In some cases, this caused damages of more than $100,000 in just one day, equivalent to around 91,000 euros.
Also of interest: Google and Samsung’s AI glasses translate conversations live
Although the configuration file of the discovered framework contains comments in Russian, Google does not attribute the tool to any specific group or country. The data on the server suggests that the operator particularly targeted French websites in the media, entertainment, and education sectors.
Major Security Flaw in Google Chrome: Update Urgently Needed
Google Closes Android Security Vulnerability Exploited by Hackers
Companies should react faster
Security experts recommend three specific measures for companies. First, they should accurately determine which of their systems are accessible via the internet. Additionally, security updates should be installed faster than before. The time between the release of a patch and its exploitation by artificial intelligence is decreasing. Furthermore, IT teams should not only look for security vulnerabilities but also specifically search for signs of ongoing attacks.
Merritt Baer summarizes the challenge: “Attackers bring speed, defenders must bring knowledge of their own environment.”
Also, Microsoft has significantly tightened its recommendation for security updates in light of the development. Instead of within two weeks, users should now install updates within three days. The company is responding to AI-supported attacks that can turn published patches into exploits more quickly.
Google expects more automated attacks
Google anticipates that cybercriminals will increasingly automate attacks using AI in the future. The entry barrier is low, while attacks can be significantly expanded. Therefore, the company considers it very likely that such frameworks will be used more frequently against organizations worldwide.
According to its own statements, Google has blocked all accounts and resources associated with the misuse of its Gemini models.