World's First AI-Driven Cyberattack Couldn't Breach OT Systems

A small, non-state hacking group conducted an AI-directed campaign between December 2025 and February 2026, using Claude Code to generate exploitation frameworks and guide attacks that led to the theft of millions of government records across at least nine Mexican entities. The attackers successfully exfiltrated tax, property, and other records from IT networks but failed to pivot into the operational network of the Monterrey water utility due to OT protections (data diode, segmentation, and authentication), leaving only limited IT-sourced loot; the case highlights how AI lowers the skill barrier for leveraging existing IT weaknesses while mature OT controls remain effective mitigations.