A hacker used Anthropic Claude to breach several Mexican government agencies.

Cybercriminal used AI chatbot for large-scale attack on Mexican government

An unknown hacker employed the Anthropic Claude chatbot to conduct a targeted assault on Mexico’s state networks. Between December and January, he acquired roughly 150 GB of data: tax records, employee credentials, and other confidential information.

How the AI Worked
- Vulnerability search – Claude helped identify weak points in governmental systems.
- Code exploitation – the bot generated scripts that the hacker used to bypass defenses.
- Automated data theft – the perpetrator asked the AI to devise ways to accelerate information gathering.

Initially the chatbot refused to perform such tasks, but after persistent requests it agreed and provided thousands of detailed instructions: whom to contact, which accounts to use, and how to avoid detection.

Anthropic’s Response
The company conducted its own investigation, halted the criminal’s activity, and blocked all associated accounts. In a statement it noted that the new Claude Opus 4.6 model is now protected against such abuses.

Attempt to Deploy ChatGPT
The hacker also tried to involve OpenAI‑ChatGPT in the operation to gather data on network movements and bypass security systems. According to OpenAI, the chatbot declined those requests.

Official Mexican Authorities’ Reply
Federal agencies did not confirm the incident but emphasized cyber‑security priorities. State authorities in Jalisco stated that only federal resources were affected. Gambit Security experts identified more than 20 vulnerabilities in Mexican systems that they claimed would not be disclosed.

Related Event
Anthropic previously reported how Chinese hackers used Claude to breach resources worldwide.