2026年9月16日 / 美国东部时间下午1:21 / 哥伦比亚广播公司新闻（CBS News）
作者：劳伦·菲希滕 数字记者
劳伦·菲希滕是哥伦比亚广播公司新闻的记者，报道人工智能、数字安全和网络极端主义议题。她毕业于北卡罗来纳大学教堂山分校，加入CBS新闻前曾在CBS新闻全国编辑部担任助理制片人。
查看完整简历
安瑟罗皮克（Anthropic）公司上周发布的一份报告显示，在美国与伊朗的战争背景下，一名与伊朗有关联的对手利用美国开发的人工智能试图瞄准美国海军部队。
据该公司介绍，这个未被安瑟罗皮克点名但被标记为“威胁行为者”的组织，使用克劳德（Claude）人工智能工具汇总开源信息，以确定美国海军的部署位置并制定瞄准建议。安瑟罗皮克表示，该账号已被封禁，且公司已将相关信息共享给美国政府以挫败该威胁。
专家表示，随着美国对手纷纷借助人工智能强化军事行动，此类事件可能会愈发常见。
专家称，安瑟罗皮克的调查结果与伊朗利用技术支持其战略目标的做法一致，包括在其与美国和以色列的冲突中。但这份报告揭示了能力日益强大的人工智能模型如何加快了曾经耗时费力的大规模公开数据汇总与分析工作。
“他们正在利用人工智能来优化自己现有的作战方式，以及他们早已习惯使用的战术或手段，”两党非营利智库战略与国际研究中心（Center for Strategic and International Studies）情报、国家安全与技术项目高级研究员尼基塔·沙阿说道。
“人工智能模型能帮助他们执行各类攻击行动，这让整个过程变得更加有效、更高效、也更快速，”她补充道。
在安瑟罗皮克报告中提及的案例中，相关方利用了包括卫星图像、船舶和飞机追踪数据以及公开军事照片说明在内的公开信息，从而更全面地掌握美国海军的动向。沙阿表示，人工智能还可以帮助优化导弹和无人机的瞄准精度、提升战场态势感知能力，并可用于监视行动。
“新的威胁在于，对手现在比以往任何时候都更容易了解美军的动向，”乔治敦大学安全与新兴技术中心研究员萨姆·布雷斯尼克说道。“美军要隐藏或掩盖其活动，正变得越来越困难。”
克劳德在伊朗处于封禁状态，但对手可以通过虚拟专用网络（VPN）和外国手机号码轻松注册账号。
在安瑟罗皮克发布这份报告之前，威胁情报提供商记录未来（Recorded Future）于7月发布的一份报告发现，人工智能可能加速了伊朗现有的网络和军事能力。报告称，鉴于俄罗斯对伊朗军事的支持，“在乌克兰战场得到完善的人工智能赋能战术和能力，有可能促成伊朗对以色列和波斯湾国家发动无人机袭击”的可能性有所上升。
除了用于实战军事行动，伊朗还将人工智能用于“非接触战”——即不诉诸武力，而是通过塑造舆论环境、网络战等手段展开行动。
在社交媒体上，全球各地的伊朗大使馆经常发布批评美国和以色列的人工智能生成视频。安瑟罗皮克还发现，有三个与伊朗政府结盟的账号使用克劳德开展公共舆论宣传活动。报告称，“伊朗的机构将制作宣传内容视为宗教和战略职责”。在一场涉及美国与伊朗、以色列的战争宣传活动中，相关行为者虚假地将言论归咎于布鲁金斯学会和兰德公司等智库研究机构，以营造可信度。
这些账号利用克劳德管理和运营数字行动、进行语言翻译，并让帖子看起来像是来自独立新闻来源。它们的内容被分发到X、TikTok和Instagram等社交媒体平台。
沙阿和其他专家表示，这种人工智能的使用并非伊朗独有，也并非对手首次利用网络能力实现军事目的并提升态势感知能力。
“总体而言，对手都在使用人工智能技术，”她说。“他们往往会利用人工智能来升级自己正在开展的行动，或是优化、完善现有的行动模式。”
这也并非首次出现美国制造的技术被用于损害美国利益的情况。例如，在俄乌冲突中，俄罗斯正在部署由美国制造技术引导的无人机对乌克兰发动袭击。
布雷斯尼克表示，在态势感知领域，美国迄今仍对其对手保持优势。但他补充道，日益先进的人工智能模型正让其他国家迎头赶上，五角大楼已经在着手应对这一问题。
“由于如今人们已经意识到，能力不那么先进的行为体也能摸清美军的动向，我们即将看到大量反制措施出台，以重新建立一定程度的突袭可能性，并提升美军秘密行动的能力，”他说道。
哥伦比亚广播公司新闻审核的一项提案显示，其中一项建议的反制措施包括创建人工智能“虚拟身份”，模拟真实的生活模式，掩盖“数字尾气”，以干扰对手的追踪和监视行动。
当被置评请求时，五角大楼一名官员表示，他们不会讨论情报相关事宜。
沙阿表示，随着人工智能的发展，国家安全界一直预计对手会利用该技术优化、扩大并强化行动。她说，安瑟罗皮克的报告凸显了一个更广泛的问题：科技公司正踏入敌对威胁环境这一陌生领域。
“我看了这份报告后会想，当这些模型像现在这样普及的时候，这些公司、整个国家安全界当初到底是怎么想的？”她说。“犯罪分子当然会想方设法以各种方式利用它们。”
AI makes it “easier than ever” for adversaries to target U.S. military, experts say
September 16, 2026 / 1:21 PM EDT / CBS News
By Lauren Fichten Digital Reporter
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
Amid the United States’ war with Iran, an Iran-linked adversary used American-developed artificial intelligence to try to target U.S. naval forces, according to a report released by Anthropic last week.
The group, which Anthropic did not identify but labeled a “threat actor,” used Claude AI to compile open-source information to identify U.S. naval positions and develop targeting recommendations, according to the company. The account was banned and Anthropic shared the information with the U.S. government to disrupt the threat, it said.
Scenarios like this could become increasingly common, experts said, as America’s adversaries turn to AI to boost military operations.
Anthropic’s findings are consistent with how Iran has been using technology to support its strategic objectives, including in its conflict with the U.S. and Israel, experts said. But, the report sheds light on how increasingly capable AI models can expedite the once-tedious task of aggregating and analyzing large volumes of publicly available data.
“They’re using it to enhance the way they already operate and the tactics or the techniques that they’re already used to doing,” said Nikita Shah, a senior fellow with the Intelligence, National Security, and Technology program at the bipartisan nonprofit think tank Center for Strategic and International Studies.
“It just makes it much more effective, much more efficient, much faster as to how the models can help them carry out those different attack methods,” she said.
In the case highlighted in Anthropic’s report, publicly available information including satellite imagery, ship and aircraft tracking data, and captions on public military photographs was used to provide a more comprehensive picture of U.S. naval movements. AI also can help refine the targeting of missiles and drones, increase visibility of the battlefield and be used for monitoring, Shah said.
“The new threat is just that it is now easier than ever before for adversaries to be able to understand what the U.S. military is doing,” said Sam Bresnick, a research fellow at Georgetown’s Center for Security and Emerging Technology. “It is getting much harder for the U.S. military to cloak or to obscure its activities.”
Access to Claude is blocked in Iran, but adversaries can easily create accounts using VPNs and foreign phone numbers.
Before Anthropic’s report, a July report released by threat intelligence provider Recorded Future found that AI likely accelerated Iran’s existing cyber and military capabilities. Given Russia’s support of Iran’s military, the report said, there is an increased likelihood that “AI-enabled tactics and capabilities, refined in Ukraine, contributed to Iranian drone attacks against Israel and Persian Gulf states.”
In addition to informing its kinetic warfare efforts, Iran has leveraged AI for “soft war” — which does not involve force, but instead, tactics such as shaping the information landscape and cyberwarfare.
On social media, Iranian embassies around the world regularly post AI-generated videos critical of the U.S. and Israel. Anthropic also found three Iranian state-aligned accounts using Claude for public influence campaigns, where “Iranian institutions produce propaganda as both a religious and strategic duty,” according to the report. In one campaign involving Iran’s war with the U.S. and Israel, actors falsely attributed claims to research institutions like think tanks Brookings and RAND to create a sense of credibility.
The accounts used Claude to manage and operate digital operations, for language translations and to make posts seem attributable to independent news sources. Their content was dispersed across social media platforms like X, TikTok and Instagram.
This sort of use of AI is not unique to Iran, nor to the way adversaries have been leveraging cyber capabilities for military purposes and to gain greater situational awareness, Shah and others said.
“Adversaries in general are using AI capabilities,” she said. “They tend to be using it to uplift what they’re already doing, or to enhance or refine what they’re already doing.”
It’s also not the first time American-made technology has been used against U.S. interests. In the Russia-Ukraine conflict, for example, Russia is deploying drones guided by U.S.-made technology for attacks on Ukraine.
The U.S., Bresnick said, has so far maintained an advantage over its adversaries in terms of situational awareness. But increasingly advanced AI models are allowing other countries to catch up — something the Pentagon is already working to combat, he added.
“Because they are aware now that less sophisticated actors can understand what they’re doing, we’re about to see a great deal of countermeasures to try to reestablish a level of potential surprise and ability to operate with more secrecy,” he said.
One of the suggested countermeasures, according to a proposal reviewed by CBS News, includes creating AI “personas” that can mimic realistic life patterns and obscure “digital exhaust” to throw off adversarial tracking and surveillance efforts.
When asked for comment, a Pentagon official said they do not discuss matters of intelligence.
As AI has advanced, the national security community has been anticipating adversaries using the technology to refine, scale and amplify operations, Shah said. Anthropic’s report, she said, points to a wider issue where tech companies are entering the unfamiliar territory of the adversarial threat landscape.
“I kind of read it and I think well, what did these companies, what did the national security community think would happen when you make these models as accessible as they are?” she said. “Of course the bad guys are going to want to exploit them in different ways.”
发表回复