Google First Reports Russian-Language Hacker AI Framework Achieving Fully Automated Exploit Deployment, Unattended Attacks on French Media and Education Sites

Google's Threat Intelligence team disclosed that Russian-speaking attackers used an AI framework called exploit_agent to complete the full chain from patch

Google's Threat Intelligence team announced on November 16, 2026, that Russian-speaking attackers used an AI framework called exploit_agent to complete the entire process from patch diff scanning, vulnerability identification, and exploit code generation to automated attack deployment, with no human intervention throughout, targeting French media, entertainment, and education websites.

Factual Reconstruction

The framework is composed of multiple AI agents, combining frontier models with locally hosted language models, continuously comparing code changes in mainstream software before and after vendors release patches. Once a patch appears, the system can infer the meaning of the change, draft exploit code, and execute it directly. Researchers also found Russian-language comments and configuration files in the operational code, instructing users to enter API keys into "auto-exploitation" rather than a chat interface. Victim data points to French sites in relevant industries.

Google's report also mentioned that after two PHP vulnerability patches were released on October 21, a large number of purported exploits appeared within days, most of which failed to run properly and were suspected to be AI-generated. During August, Google tracked 22 exploited zero-day vulnerabilities, higher than the monthly average of 11 in early 2026 and 8 in 2025.

Mechanism Breakdown

exploit_agent links vulnerability research with intrusion execution, forming a closed loop. The system not only automatically generates exploits but also includes scripts for internal reconnaissance, password guessing, and installing webshells, some of which remove competing attackers' webshells. This differs from previous cases where AI was mainly used to write phishing emails or summarize stolen data, marking a shift from "accelerating human operations" to "unattended pipelines."

The report also noted that open-source attack frameworks CAIRN and PentAGI have been observed used in activities targeting U.S. aerospace organizations and Asian institutions, while China-linked groups are building similar automated attack frameworks.

Industry Impact

Low barriers to entry, ease of integrating new exploits, and potential for large-scale expansion are expected to lead threat actors to continue adopting such frameworks, affecting organizations worldwide. The rapid emergence of AI-generated exploits after vendor patch releases increases pressure on defenders' response windows.

Strategic Assessment (Analysis, Not Fact)

From current observations, the emergence of fully automated attack frameworks extends compliance and treaty obligations from model overreach issues to the actual deployment of weaponized agents. In historical precedents, AI was mostly used as an auxiliary tool to accelerate manual processes, but this incident shows that autonomous chains have formed, and defense strategies need to adjust monitoring and blocking capabilities against unattended attack pipelines accordingly.

The report emphasizes that the low-barrier nature of such tools may accelerate activities targeting various organizations worldwide, and relevant parties need to monitor the proliferation of open-source frameworks and framework-building activities by state-linked groups.