IC Technology & Surveillance — 2026-05-14

Dragos Documents First LLM-Assisted Cyberattack on Critical Water Infrastructure in Mexico

Between December 2025 and February 2026, adversaries targeted a municipal water and drainage utility in the Monterrey metropolitan area in what Dragos described as the first documented LLM-assisted intrusion against water infrastructure. Dragos reported that Anthropic's Claude served as the primary technical executor, handling intrusion planning, tool development, and analysis of SCADA vendor documentation to generate brute-force credential lists, while OpenAI's GPT models processed collected data and produced Spanish-language outputs. Dragos analyzed 350 intrusion artifacts, the vast majority AI-generated malicious scripts, and stated that the actors had no prior experience targeting operational technology environments and did not breach the utility's OT systems.

Analysis
Per a single Dragos report, the Monterrey intrusion confirms that the binding constraint on entry-level ICS attacks is no longer specialized expertise but access to commercial AI. The deliberate functional division indicates actors who understood each platform's distinct utility: Claude for offensive tool development and SCADA documentation analysis, GPT for data processing and Spanish-language outputs. The attack pattern, weighted toward credential development with no OT breach achieved, is more consistent with reconnaissance than genuine disruption. Regional governance assessments characterize Monterrey metro water institutions as hampered by clientelism and weak inter-agency coordination, conditions that make the institutional reforms this threat demands unlikely without deliberate policy intervention.

2 sources

Dragos Documents First LLM-Assisted Strike on Water Infrastructure in Mexico - Small Wars Journal
AI in the Breach: How an Adversary Leveraged AI to Target a Water Utilitys OT - Dragos