Rubicon crossed: first cyberattack carried out by autonomous AI agent identified

A report by Google and Mandiant has for the first time documented a cyberattack fully executed by an autonomous AI agent named exploit_agent, allegedly used by Russian hackers. The system independently hacked French websites. The report also notes Iran's use of AI for intelligence and information operations.

The quarterly report by Google Threat Intelligence Group (GTIG) and Mandiant, published in mid-November, documented the first cyberattack fully executed by an autonomous AI agent. The exploit_agent system, allegedly used by Russian hackers, independently scanned networks, identified vulnerabilities, wrote exploit code, and attacked French websites in the communications, entertainment, and education sectors. The report also notes that the Iranian group CALANQUE ION (APT42) uses language models for intelligence and social engineering, and creates fake digital personas for information operations in the Middle East and Europe. In response, Google is developing autonomous defense systems Big Sleep and CodeMender. The number of exploited vulnerabilities nearly doubled from January to August 2026, and zero-day vulnerabilities reached a record 22 cases in August. Google's lead analyst John Hultquist warned that traditional response methods are no longer effective against machine-speed attacks.

Rubicon crossed: first cyberattack carried out by autonomous AI agent identified