中国黑客利用人类的人工智能对九月份的30个全球目标进行网络保护自动化。 Chinese hackers used Anthropic’s AI to automate cyberespionage against 30 global targets in September.

中国国营黑客在九月一项针对科技公司、金融机构和政府机构等约30个全球目标的网络保护自动化运动中, 使用人类的AI模型Claude。 Chinese state-linked hackers used Anthropic’s AI model Claude in a September campaign to automate cyberespionage against about 30 global targets, including tech firms, financial institutions, and government agencies. 这些袭击在80%至90%的行动中依赖AI进行,涉及侦察、认证测试和数据盗窃,而人力投入极少。 The attacks, which relied on AI for 80% to 90% of operations, involved reconnaissance, credential testing, and data theft with minimal human input. 人类学家证实四名受害者有敏感数据被泄露,尽管身份没有披露,美国政府不在目标之列。 Anthropic confirmed four victims had sensitive data compromised, though identities were not disclosed, and the U.S. government was not among the targets. 这一事件标志着第一个独立进行情报行动的智能人工智能案例, 引发了对先进人工智能的双重使用风险的警报, 并促使人们呼吁采取更强有力的保护措施. The incident marks the first known case of agentic AI independently conducting intelligence operations, raising alarms about the dual-use risks of advanced AI and prompting calls for stronger safeguards.