Claude AI used in attempted compromise of Mexican water utility

Researchers documented an unknown cyber threat group using Anthropic's Claude to facilitate a sophisticated takeover attempt against a Mexican water utility. Claude rapidly interpreted unfamiliar OT/ICS infrastructure and developed plausible access paths without prior ICS-specific context — concrete evidence that frontier LLMs are accelerating non-expert attackers into critical-infrastructure territory.