French websites in the media agentic AI attack — Nov 2026
Google researchers identified an autonomous cyberattack campaign where Russian-speaking hackers used an AI system to exploit software vulnerabilities without.
- Reported date
- Nov 18, 2026
- Target
- French websites in the media, entertainment and education sectors
- Agent type
- Other AI agent
- Agent role
- Used by the attacker
The exact incident date was not established. This entry is dated by its source report.
Google’s Threat Intelligence Group and Mandiant have identified what they characterize as the first end-to-end cyberattack executed entirely by artificial intelligence. The operation, which targeted French websites in the media, entertainment, and education sectors, represents a shift from AI-assisted hacking to fully autonomous intrusion pipelines.
What happened
Researchers discovered an autonomous system dubbed "exploit_agent" that operates without human intervention. The system utilizes a chain of AI agents, incorporating both frontier and locally hosted language models, to monitor software vendors for new patches. By comparing code before and after a patch is published, the system infers the nature of the vulnerability, drafts an exploit, and executes the attack automatically.
Evidence of the operation was found on a staging server containing victim data. Investigators attributed the activity to Russian-speaking hackers based on comments found within the code, including configuration files that instructed users to input API keys for "auto-exploitation." Beyond the primary exploit generation, the system included scripts designed for internal reconnaissance, password guessing, and the installation of webshells. Notably, the system also contained functionality to identify and erase webshells belonging to rival attackers.
Google analysts highlighted this incident as a critical inflection point in cyber warfare. While previous adversary use of AI was limited to accelerating human-directed tasks—such as drafting phishing lures or summarizing stolen data—this campaign demonstrates a fully unattended pipeline from vulnerability research to system compromise. Google researchers warned that the low barrier to entry, combined with the ability to integrate new exploits and scale operations, suggests that threat actors will likely continue to deploy such frameworks in global campaigns. The report emphasizes that the primary advantage of this autonomous approach is the speed at which it can weaponize newly disclosed vulnerabilities.
Evidence in the reporting
- Incident evidence
- Russian-speaking hackers carried out what its analysts believe is the first
- Agent involvement
- it chains together multiple AI agents with frontier and locally hosted