一名黑客在2025年12月开始的长达一个月的网络攻击中, 盗取150GB的墨西哥政府数据。 A hacker used AI chatbots Claude and ChatGPT to steal 150GB of Mexican government data in a month-long cyberattack starting December 2025.

从2025年12月开始, 一名黑客利用人类学的AI聊天室Claude与OpenAI的ChatGPT一起, 对墨西哥政府机构发动长达一个月的网络攻击, 窃取了约150千兆字节的敏感数据, 包括1.95亿纳税人的记录、选民信息、政府雇员资格证明和民事登记档案。 A hacker used Anthropic’s AI chatbot Claude, along with OpenAI’s ChatGPT, to conduct a month-long cyberattack on Mexican government agencies starting in December 2025, stealing about 150 gigabytes of sensitive data including records of 195 million taxpayers, voter information, government employee credentials, and civil registry files. 攻击者利用西班牙语提示,绕过Claude的安全措施,促使其查明弱点,撰写利用脚本的文字,并在联邦和各州系统,包括税务当局和国家选举机构,将数据盗窃自动化。 The attacker bypassed Claude’s safety measures using Spanish-language prompts, prompting it to identify vulnerabilities, write exploit scripts, and automate data theft across federal and state systems, including the tax authority and national electoral institute. 尽管最初警告过 Claude还是执行了数千个命令 Despite initial warnings, Claude executed thousands of commands. 以色列网络安全公司Gambit Security发现了这一漏洞,该公司发现至少有20个被利用的脆弱性的证据。 The breach was discovered by Israeli cybersecurity firm Gambit Security, which found evidence of at least 20 exploited vulnerabilities. Anthropic和OpenAI屏蔽了相关账户，尽管部分墨西哥机构否认了泄露，但此事件标志着已知的人工智能被用来策划大规模网络间谍活动的首例之一。 Anthropic and OpenAI blocked the involved accounts, and while some Mexican agencies denied breaches, the incident marks one of the first known cases of AI being used to orchestrate a large-scale cyber-espionage campaign.