2026年9月10日 美国东部时间下午3:49 / 哥伦比亚广播公司新闻（CBS News）
作者：劳伦·菲希滕（Lauren Fichten）
劳伦·菲希滕是哥伦比亚广播公司新闻负责人工智能、数字安全与网络极端主义报道的记者。她毕业于北卡罗来纳大学教堂山分校，此前曾在哥伦比亚广播公司新闻全国编辑部担任助理制片人。
查看完整简介
Anthropic公司周四表示，其阻止了以“可能支持生物武器研发”的方式使用其Claude AI模型的科学家。
这一披露来自一份长篇报告，该报告同时揭露了其他涉及监视、诈骗、常规武器研发和宣传的有害活动。该公司称，报告中的案例是“Anthropic迄今发现的最显著、最新型的威胁活动”实例。报告指出，威胁行为者包括犯罪分子、疑似受国家支持的组织、间谍软件供应商以及国家宣传机构。
Anthropic分享了五起涉及可能支持生物武器研发活动的案例。该公司将生物用途滥用称为“前沿AI模型最严重的风险之一”。
该公司表示，所发现的案例“提供了AI模型具备相关能力的证据”，但“无法具体证明这种能力会在现实世界中被用于研发生物武器”。
这些案件涉及的人员均为在职科学家，Anthropic选择不公开他们的身份。该公司表示，无法“断言他们有意造成伤害”。
报告中称：“当我们发现并调查这些案件时，我们封禁了这些用户的账号，并将调查结果纳入我们的前沿模型 safeguards、执法和威胁情报流程，以更好地在未来预防、发现和阻止此类活动。”
报告称，近期推出的更先进的模型如Claude Fable 5比早期模型拥有更强的防护措施，会限制用户访问“广泛的两用生物研究查询”，并指出AI的生物能力既可用于有益用途，也可用于有害用途。
blob:https://www.cbsnews.com/56669ae3-69b2-4ba5-8fe0-c8f7b4c5befc
该公司表示：“同样的信息既可用于研发生物武器，也可用于研发例如疫苗或疾病治疗方法。”
宣传与监视
Anthropic还表示，其发现并移除了利用Claude开展影响公众舆论的宣传行动的账号，其中包括三个与伊朗政府结盟的账号。
该公司称，这些行动均由伊朗国家宣传机构内部或代表该机构的行为者主导。他们使用Claude制作内容，让帖子看起来像是来自独立新闻来源，并在X、Instagram和TikTok等社交媒体平台上传播内容。
一名与伊朗相关的威胁行为者利用Claude制定了针对该地区美国海军部队的目标推荐方案。该账号还为伊朗国家系统的国内大规模监视平台设计了软件。
中国和西非的行为者还利用Claude搭建和开展监视行动，在某些情况下还用于识别目标。Anthropic表示，其发现了一些案例，其中AI被用来替代工程劳动力。
在中国，三个与中国市级安全部门有关联的账号利用Claude开展监视和跨国镇压行动，其中一个市级部门负责对海外活动人士和组织进行身份识别。
Anthropic says it disrupted scientists using Claude AI for possible biological weapons development
September 10, 2026 3:49 PM EDT / CBS News
By Lauren Fichten
Lauren Fichten is a journalist at CBS News covering artificial intelligence, digital safety and online extremism. She joined CBS News after graduating from UNC-Chapel Hill and was previously an associate producer at the CBS News National Desk.
Read Full Bio
Anthropic said Thursday it blocked scientists who used its Claude AI models “in ways that could support biological weapons development.”
The revelation was shared in a lengthy report that also divulged other harmful activity involving surveillance, scams, conventional weapons development and propaganda. The cases in the report are examples of “the most notable and novel threat activity” identified by Anthropic so far, the company said. Threat actors included criminals, suspected state-sponsored groups, spyware vendors and state propaganda institutions, the report said.
Anthropic shared five cases involving activity that had the potential to support the development of biological weapons. It called biological misuse “one of the most serious risks of frontier AI models.”
The company said the cases identified “provide evidence of capability” of the AI models, but “cannot concretely demonstrate that such capability would ever be used to develop biological weapons in the real world.”
The people involved in these cases are working scientists, whom Anthropic chose not to identify. The company said it can’t “assert that they intended harm.”
“When we detected and investigated these cases, we banned the users’ accounts and incorporated our investigative findings into our frontier model safeguards, enforcement, and threat intelligence processes to better prevent, detect, and disrupt these activities in the future,” the report says.
Recent, more advanced models like Claude Fable 5 contain stronger safeguards than earlier models and restrict access to “a wide range of dual-use biological research queries,” the report says, noting that the biological capabilities of AI can be used both for beneficial and harmful purposes.
blob:https://www.cbsnews.com/56669ae3-69b2-4ba5-8fe0-c8f7b4c5befc
“The same information that can be used to develop a biological weapon could also be used to develop, for example, a vaccine or a cure for a disease,” the company said.
Propaganda and surveillance
Anthropic also said that it identified and removed accounts using Claude for influence operations campaigns to shape public opinion, including three Iranian state-aligned accounts.
In these cases, each operation was run by an actor working within or on behalf of an Iranian state propaganda institution, the company said. Claude was used to build content, make posts seem like they were from independent news sources and to proliferate content across social media platforms like X, Instagram and TikTok.
One Iran-nexus threat actor used Claude to develop targeting recommendations against U.S. naval forces in the region, Anthropic said. The same account designed software for a domestic mass-surveillance platform for Iranian state systems.
Claude was also used to build and run surveillance operations by actors in China and West Africa, and in some cases, to identify targets. Anthropic said it identified cases where AI was being used in place of an engineering workforce.
In China, three accounts aligned with the PRC municipal security service used Claude for surveillance and transnational repression, including a municipal bureau that profiles overseas activists and organizations.
发表回复