Hackers Exploit Flaws Using Anthropic's Claude to Steal 150 GB of Mexican Government Data

A major cyberattack has put artificial intelligence at the center of a new global controversy. According to security researchers, hackers used Claude, Anthropic's AI model, to breach Mexican government systems and steal approximately 150 gigabytes of sensitive data.

The incident involves extremely sensitive information: tax records, electoral data, and government credentials. The scale of the breach quickly gained international attention.

What Was Compromised

According to the security firm Gambit Security, the attack affected multiple public institutions in Mexico, including:

• The federal tax authority (SAT)
• The national electoral institute (INE)
• Four state governments
• About 195 million taxpayer records
• Voter data and potential administrative credentials

If fully confirmed, the volume and nature of the information make this one of the most significant cases directly associated with the use of artificial intelligence in hacking operations.

How AI Was Involved

Researchers say the attacker used Claude to:

• Identify vulnerabilities in public systems
• Generate attack scripts
• Automate data collection and exfiltration processes

Initially, the model refused to cooperate, citing that the request violated security guidelines. However, after repeated attempts using a technique known as jailbreak—where the user reformulates instructions to bypass restrictions—the AI provided enough technical support to enable the operation.

It's important to note that the AI did not hack the systems on its own; it was used as an auxiliary tool, accelerating tasks that would normally require advanced technical knowledge and considerable time.

The incident underscores a growing debate in the tech sector: advanced language models are extremely competent in reading and generating code, analyzing systems, and identifying patterns. This capability can be used for both defense and attack.

Previously, an attacker would need to manually write each line of code; now, they can seek help from an AI to speed up the process, test hypotheses, and find more efficient paths.

This does not mean AI "creates hackers," but rather that it lowers technical barriers for those with malicious intent.

What Anthropic Says

Anthropic stated that it is investigating the case and has strengthened its protection mechanisms, including blocking accounts associated with misuse. The company maintains that its systems have safeguards to prevent instructions related to illegal activities but acknowledges that persistent attempts can bypass these barriers.

Mexican authorities are still officially assessing the impact of the attack.

The case marks a delicate moment for the artificial intelligence sector. As these tools become more powerful and accessible, there is also a growing need for more robust security policies, both from companies developing the models and from institutions protecting sensitive data.