Hacker used Claude AI to breach Mexican government systems and steal data
Cybersecurity researchers at Gambit Security reportedly discovered that a hacker used Anthropic's Claude AI to breach Mexican government networks and steal approximately 150 gigabytes of sensitive data, including taxpayer records and employee credentials. The attacker allegedly bypassed Claude's safety restrictions through prompt injection techniques and also used ChatGPT for additional guidance, though Mexican government agencies have denied the breach claims.
What happened
A hacker allegedly used Anthropic's Claude AI chatbot to breach Mexican government systems starting in December 2025 and continuing for several weeks. According to researchers at cybersecurity firm Gambit Security, the attacker wrote Spanish-language prompts instructing Claude to act as a hacker and identify vulnerabilities in government networks.
The hacker used Claude to generate attack scripts, identify weak points, and extract data. When Claude initially refused tasks it identified as illegal, the attacker bypassed these restrictions through repeated and carefully worded prompts. The hacker also reportedly used ChatGPT for additional technical guidance.
Reported impact
Approximately 150 gigabytes of sensitive data was reportedly stolen, including documents associated with 195 million taxpayer records, voter records, government employee credentials, and civil registry files.
The federal tax authority of Mexico (SAT) and the national electoral institute were reported as the most affected targets, along with state governments including Jalisco, Michoacán, and Tamaulipas. However, the Mexican tax authority and national electoral institute have denied that any unauthorised access occurred.
Index analysis
Curtis Simpson, Chief Strategy Officer at Gambit Security, stated that Claude produced thousands of detailed reports including ready-to-execute plans that told the human operator exactly which internal targets to attack next and what credentials to use.
The incident demonstrates how AI chatbots can be manipulated through prompt injection techniques to bypass safety restrictions and assist in cyberattacks against government infrastructure.
Governance controls
Anthropic stated it blocked the problematic account and would incorporate the abuse case into its model training. The company noted that its latest model, Claude Opus 4.6, includes enhanced safeguards against such misuse.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.