Hacker Uses Anthropic AI To Steal Sensitive Information From The Mexican Government

A hacker managed to steal sensitive tax and voter information from the Mexican government by using Anthropic's AI chatbot Claude, according to a new report.

Israeli cybersecurity company Gambit Security wrote in research that an unknown user tasked the chatbot with acting as an elite hacker to do so. Researches said they don't believe the hacker is tied to a foreign government.

The hacking began in December and continued for about a month. It managed to steal 150 gigabytes of data, including taxpayer and voter records, as well as employee credentials and civil registry files. The water utility of the state of Monterrey was also compromised.

Claude warned the user about the consequences of such actions but ended up complying with the requests and conducted the hacking.

Anthropic said it investigated the claim, disrupted the activity and banned the accounts involved. A company representative said one of the latest Claude models includes probes that can disrupt misuse.

Mexican officials seemed to react to the development in December, saying they were investigating the breaches. It was not clear whether they were referring to this incident in question.

Bloomberg noted that researchers at Amazon said last week that a small group of hackers managed to break into over 600 firewall devices across dozens of countries by using available AI tools.