A serious cybersecurity incident involving AI has put Mexico’s government networks under the spotlight, after hackers allegedly used an AI chatbot to pull off a large-scale data theft. According to a Bloomberg investigation, the attacker relied on Anthropic’s Claude AI to break into official systems, stealing close to 150GB of sensitive data that reportedly included taxpayer details, internal documents and employee login information.
Hackers attack Mexico govt using Claude AI, steal 150GB data: What really happened?
The breach began quietly in December and went on for several weeks before being detected. Cybersecurity firm Gambit Security, which tracked the activity, said the hacker leaned heavily on AI to do the heavy lifting. Instead of manually probing systems, the attacker used Claude to identify weak points in government networks, generate attack scripts and even streamline the process of extracting data.
What alarmed researchers most was how the chatbot was pushed beyond its safety limits. Claude initially refused to help with tasks that were clearly illegal. However, through repeated and carefully worded prompts, the hacker appears to have bypassed those restrictions, effectively forcing the AI to assist in planning and executing the attack.
“In total, it produced thousands of detailed reports that included ready-to-execute plans, telling the human operator exactly which internal targets to attack next and what credentials to use,” said Curtis Simpson, chief strategy officer at Gambit Security.
Anthropic has since confirmed that it looked into the incident. The company said it shut down the activity and banned all accounts connected to the misuse. It also pointed out that its latest AI model, Claude Opus 4.6, includes stronger safeguards aimed at preventing this kind of abuse.
Investigators say Claude was not the only AI tool involved. The hacker also reportedly used ChatGPT, developed by OpenAI, to gather additional technical knowledge. This included understanding how to navigate complex computer networks, figuring out access requirements and learning methods to stay under the radar. OpenAI has said it detected attempts to break its rules and that its systems blocked such requests.
The person behind the attack has not been identified, and no group has officially claimed responsibility. While there is no confirmed attribution, Gambit Security has suggested the operation could be linked to a foreign state actor. The motive also remains unclear, raising concerns about how the stolen data could eventually be used.
Official responses from Mexico have been limited. The country’s national digital agency has not directly addressed the breach but has reiterated that cybersecurity is a key focus area. The state government of Jalisco has denied that its systems were compromised, maintaining that the incident affected only federal networks. Mexico’s national electoral institute has also said it has not seen any unauthorised access or breaches in recent months.
At the same time, Gambit Security said its research uncovered at least 20 major security gaps across government systems, suggesting that the full picture may be more complicated than what has been publicly acknowledged.
The episode adds to growing concerns around the misuse of advanced AI tools. Claude has previously been linked to other hacking campaigns, including attempts last year to infiltrate multiple international targets.
