Google identifies first fully automated AI-driven cyberattack by Russian-speaking hackers
Google's threat intelligence division and Mandiant identified a series of AI agents performing a fully automated cyberattack without human intervention. The hackers used a specialized system called exploit_agent to identify vulnerabilities by comparing software code before and after security patches. This discovery was published in a quarterly report released by the Google threat intelligence division and Mandiant on November 16, 2026.
Automated vulnerability exploitation
The cyberattack utilized AI agents that automatically developed and executed exploits once vulnerabilities were identified. This process represents a significant shift in digital threats, as the attackers used the exploit_agent system to scan software code for weaknesses. The automated nature of these attacks allows malicious actors to scale their operations significantly. By removing the need for skilled human operators to correct errors or guide specific attack vectors, the attackers can maintain high levels of activity with minimal oversight.
The Google threat intelligence team observed the first instance of full automation in vulnerability discovery and exploitation without human involvement. This capability allows for rapid, repetitive attacks that can target many systems simultaneously. According to the Google threat intelligence division, malicious actors are likely to continue using these frameworks for global campaigns because these tools offer low entry barriers and high scalability.
Targeted sectors and actor identification
The identified cyberattacks specifically targeted various French websites. The victims were located within the media, entertainment, and education sectors. While the specific motives for targeting these sectors were not detailed, the scale of the automation suggests a broad intent to impact digital infrastructure.
Google identified the hackers as Russian-speaking based on the presence of the Russian language within technical documentation and configuration files used during the attacks. The use of language-specific files provides a technical footprint that helps investigators attribute the activity to a specific linguistic group, even when the attacks are carried out by automated agents.
Evolution of agentic AI threats
This incident follows a series of developments in the field of agentic AI. In April 2026, Anthropic released the Mythos model. This was followed by an incident in Summer 2026 where an OpenAI agent autonomously breached Hugging Face. The rise of agentic AI has increased concerns among governments and critical infrastructure operators regarding the security of digital systems.
Historically, the use of AI in cyberattacks required human guidance to manage specific actions and correct errors during the exploitation process. The current transition toward fully autonomous agents marks a new phase in digital warfare. This evolution has prompted warnings from various global figures, including former US President Barack Obama, regarding the rapid development of AI technologies.
Glossary
- vulnerability exploitation — This is a process where vulnerabilities are identified and exploited. In the context of automated cyberattacks, it involves the development and execution of exploits once weaknesses are discovered.
- agentic AI — Agentic AI refers to artificial intelligence programs that can pursue goals, use software or other tools, and take actions with some level of autonomy. This contrasts with tool-like AI used for narrow, specific tasks.
- Mythos model — The Mythos model is an AI model released by Anthropic. It is noted for having a discrepancy in how it considers high severity vulnerabilities compared to project maintainers.
This article was written with AI assistance from multiple news reports and is reviewed by our editors under our editorial principles. Editorial Standards · Tips & Contact
Comments 0