Article By Lucas Nolan
A hacker successfully used Anthropic’s Claude AI to conduct attacks on multiple Mexican government agencies, resulting in the theft of approximately 150 gigabytes of sensitive official data including tax records.
Bloomberg reports that a hacker manipulated Anthropic’s Claude AI chatbot to identify vulnerabilities in Mexican government networks and execute attacks that compromised taxpayer records, employee credentials, and other sensitive information.
The cyberattack, which began in December and continued for approximately one month, demonstrated how AI tools can be weaponized despite built-in safety measures. The hacker utilized Claude to locate security weaknesses in government systems, write exploitation scripts, and develop automated methods for data extraction, according to findings from cybersecurity firm Gambit Security.
The perpetrator reportedly succeeded in bypassing Claude’s protective guardrails through carefully crafted prompts. While Claude initially rejected the malicious requests, the chatbot eventually complied with the hacker’s demands.
Curtis Simpson, chief strategy officer at Gambit Security, described the extent of Claude’s involvement in the operation. “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,” Simpson said.
Following the discovery of these attacks, Anthropic launched an investigation into the incident. A company representative confirmed that the organization has disrupted the malicious activity and permanently banned all accounts associated with the breach. The spokesperson also noted that Anthropic’s newest model, Claude Opus 4.6, incorporates enhanced tools specifically designed to prevent this type of misuse.
The hacker apparently employed multiple AI platforms during the operation. Reports indicate that OpenAI’s ChatGPT was also used to supplement the attacks, with the perpetrator leveraging the chatbot to gather intelligence on navigating computer networks, identifying necessary credentials for system access, and implementing detection avoidance techniques. OpenAI confirmed it identified attempts by the hacker to violate its usage policies and stated that its systems refused to comply with the inappropriate requests.
The identity of the hacker remains unknown, and the attacks have not been definitively linked to any specific organization. However, Gambit Security suggested the possibility of foreign government involvement. The ultimate purpose for the stolen data also remains unclear.
Responses from Mexican authorities have been mixed and somewhat contradictory. Mexico’s national digital agency has not issued a statement regarding the breach but has indicated that cybersecurity remains a top priority. The state government of Jalisco denied experiencing any breach, claiming that only federal networks were affected. Meanwhile, Mexico’s national electoral institute has denied any recent breaches or unauthorized access incidents. Gambit’s research uncovered at least 20 security vulnerabilities in Mexican government systems, which the country may be reluctant to publicly acknowledge.
This incident is not the first time Claude has been implicated in major cyberattacks. In a previous case last year, hackers operating from China successfully manipulated the tool to attempt infiltration of numerous global targets, achieving success in several instances.
Be the first to comment