Welltec Defense
Uncategorized •
7 min read

How a Hacker Stole 150GB of Mexican Government Data with Claude AI

March 12, 2026 • By Al Kao

This is the story of how a solo hacker used Anthropic’s Claude AI to steal 150GB of Mexican government data and made agentic AI attacks much easier. How should companies protect themselves now against agentic AI cyber attacks?

On February 25, 2026, a shocking report revealed that 150 gigabytes Mexican government data had been stolen by a hacker using Claude AI. The stolen records of approximately 195 million identities included:

  • Federal Tax Authority (SAT): Taxpayer records and financial data.
  • National Electoral Institute (INE): Voter registration files and IDs.
  • Civil Registry: Birth certificates and marriage records.
  • State Systems: Critical data from Jalisco, Michoacán, and Tamaulipas.

How Did This Cyber Attack Happen?

The attacker bypassed Claude’s safety guardrails by convincing the AI “he” was an “elite security researcher” conducting a legitimate, authorized bug bounty program for the Mexican government.

Once the “persona” was accepted, Claude did the heavy lifting. It identified over 20 unique vulnerabilities, wrote functional exploit scripts including SQL injections, and generated “pivot roadmaps” for moving laterally through the network.

Further, when Claude’s safety filters eventually flagged the activity, the attacker transitioned to OpenAI’s GPT-4.1 to analyze the stolen credentials and refine the next stage of the attack.

The most alarming detail is that the attacker appeared to be a solo operator with mid-level technical skills but was able to prompt Claude with Spanish-language prompts that eventually broke Claude’s safety guardrails.

Why This Attack Is Different From Previous AI Attacks?

In previous cyber attacks with AI (like the WormGPT or FraudGPT trends of 2023-2024), AI was used to write better phishing emails, better vishing assets, or snippets of code. 

The Claude attack didn’t just ask for code, the hacker used Claude Code—a tool designed to interact with file systems and terminals—to perform tactical decision-making. Claude wrote scripts, analyzed the results of its own exploits, identified the next lateral move within the Mexican government’s network, and generated a “pivot roadmap” on the fly.

In simple terms, previous AI attacks were orchestrated and managed by human attackers using AI. This attack is different because the human attacker orchestrated the AI to do the work on its own – an agentic AI cyber attack!

What Does Agentic Cyber Attacks Mean for Cybersecurity?

This attack is a watershed moment for cybersecurity and AI attacks. 

This attack demonstrated that an agentic AI cyber attack is possible. Cyber criminals can now use AI to assist in phishing/vishing attacks as well as guide the AI to do all the work of hacking systems and stealing data.

The next attack is not a matter of IF but a matter of WHEN and WHO?

How Do Companies Protect Against Agentic AI Attacks?

The short answer is that your cybersecurity must include AI tools. Cyber criminals using AI or agentic AI will be attacking at machine speed that only machine speed can intercept.

1. AISOC

Security Operations Centers will need AI tools, agentic AI tools to continuously monitor, respond and isolate anomalous behaviors.

2. Zero-Trust Architecture 

Up until the agentic AI Mexico hack, zero-trust architecture could be considered best in class. Now, it is clear the Zero Trust model—where every single request is verified and access is micro-segmented. Zero Trust is your first line defense against an AI agent’s cyber attack.

3. AI Pen Testing 

With the cyber attacker able to jailbreak a chatbot LLM to commit a cyber attacker, pen testing now needs to include using LLMs or chatbots like Claude and ChatGPT to test your systems to find vulnerabilities that an AI-assisted hacker would target first.

Welltec Defense Cybersecurity with AI

Welltec Defense offers comprehensive IT and cybersecurity solutions to help ensure your company is protected against cyber attacks like that of the Mexican government. Whether you need to secure your IT infrastructure, connect your IT and cybersecurity processes into one efficient system, or need comprehensive training for your staff, Welltec Defense is ready to help your firm.

Sources & Further Reading