Anthropic公司发布威胁情报报告，披露在过去八个月内已识别并阻止了针对其Claude AI模型的多起恶意使用企图。12报告指出存在五起案例涉及行为体利用其模型支持生物武器开发，12该公司将此列为"前沿AI模型最严重的风险之一"。2此外，报告还记录了六起Claude被用于开发传统武器软件的案例，包括枪支、导弹、武装无人机、炸弹和其他弹药。12
也门武装组织曾尝试利用Claude开发导弹制导软件。8Anthropic还发现来自俄罗斯和中国针对其模型的恶意活动。3使用者采取了规避和混淆手段来绕过安全防护，6其中包括来自俄罗斯、中国和伊朗等被禁止访问其模型国家的用户。6报告进一步揭示，伊朗关联方分析了推文并监控反对派账户，中国关联行为体则利用Claude起草信息以招募特定人员。5
Google也发现有人试图使用其Gemini模型获取"合成武器化生物剂的完整分步技术指南"。12Anthropic威胁情报负责人Jacob Klein表示这是"一个极其微妙的情况"。12
Anthropic has released a comprehensive threat intelligence report detailing multiple instances in which actors attempted to exploit its Claude AI models for malicious purposes, including biological weapons development.12 The company identified and blocked five cases involving the use of its models to support bioweapon development over an eight-month period from December 2025 through August 2026.12 Beyond biological threats, the report documented six additional cases in which Claude was used to develop conventional weapons software, including firearms, missiles, armed drones, bombs, and other munitions.12
The malicious activity involved a diverse range of actors spanning state-sponsored organizations, criminal networks, and propaganda agencies.2 Users employed evasion tactics and obfuscation to circumvent safety protections, with some originating from Russia, China, and Iran—countries with restricted access to the model.6 Specific incidents included efforts by a Yemeni group to develop missile guidance software using Claude78 and operations by Iran-linked entities that analyzed social media content to identify opposition accounts for surveillance.5 In another case, a Chinese municipal security actor used Claude to gather location information about overseas protest activities in Vancouver, Canada.5 Additionally, Google discovered attempts to use its Gemini model to obtain "complete step-by-step technical guidance for synthesizing weaponized biological agents."12
Jacob Klein, Anthropic's head of threat intelligence, characterized the situation as "extremely delicate,"12 underscoring the acute challenges posed by advanced AI systems. The company emphasized that biological weapons development represents "one of the most serious risks" associated with frontier AI models.2 These findings have prompted broader policy discussions, with U.S. Senator Bernie Sanders proposing legislation to ban artificial superintelligence and suspend advanced AI development.12 Meanwhile, OpenAI's chief scientist Jakub Pachocki has called for the industry to implement "voluntary slowdowns" until safeguards are established.1
评论
还没有评论，欢迎留下第一条。