Government Data Stolen After Hacker Jailbreaks Claude AI to Write Malicious Exploit Code

A reported month-long campaign (Dec 2025–Jan 2026) saw an attacker 'jailbreak' Anthropic’s Claude with persistent Spanish-language prompts—then pivot to ChatGPT—to generate functional exploit code, attack plans, and exfiltrate roughly 150GB of sensitive data from multiple Mexican federal and state agencies (including an alleged 195 million taxpayer records); the operation targeted legacy, unpatched systems and underscores the rising threat of AI-orchestrated attacks, though some affected parties have denied the incident.