Meta Tapped a Pentagon Supplier to Prototype Face Recognition for Its Glasses
Meta Tapped a Pentagon Supplier to Prototype Face Recognition for Its Glasses
Meta 聘请五角大楼供应商为其智能眼镜开发人脸识别原型
Meta is testing face-recognition software built by a company that sells surveillance tools to police departments and the United States military, as it explores bringing the technology to its smart glasses, WIRED has learned. 据《连线》（WIRED）获悉，Meta 正在测试一款由某公司开发的人脸识别软件。该公司主要向警察部门和美国军方出售监控工具，而 Meta 此举旨在探索将该技术引入其智能眼镜。
The arrangement is documented in a software license, obtained by WIRED, that was issued by Rank One Computing—a Denver-based company that derives roughly 80 percent of its revenue from government clients—and is tied to a test version of the Meta AI app that powers Meta’s Ray-Ban and Oakley smart glasses. 《连线》获得的一份软件许可协议记录了这一合作。该协议由总部位于丹佛的 Rank One Computing 公司签发，该公司约 80% 的收入来自政府客户。该许可与 Meta AI 应用程序的测试版本相关联，该应用为 Meta 的雷朋（Ray-Ban）和欧克利（Oakley）智能眼镜提供支持。
Rank One’s face recognition has been bought by the US Marshals Service, which uses it to confirm prisoners’ identities without fingerprinting them during transport, and by the Naval Criminal Investigative Service—the Navy’s police force—which purchased the company’s video tool, ROC Watch. Rank One developed long-range face recognition for US Special Operations Command under a government research contract, saying its software could identify a face from as far as a kilometer away. Police departments across the country use its algorithms too, embedded in tools they buy from other vendors. Rank One 的人脸识别技术已被美国法警局（US Marshals Service）采购，用于在押送囚犯时无需指纹即可确认其身份；海军刑事调查局（海军警察部队）也购买了该公司的视频工具 ROC Watch。根据一份政府研究合同，Rank One 还为美国特种作战司令部开发了远程人脸识别技术，声称其软件可以在一公里外识别出人脸。全国各地的警察部门也在使用其算法，这些算法被嵌入在他们从其他供应商处购买的工具中。
The license is the first known evidence of a business relationship between Meta and Rank One, and it offers a rare look at the kind of technology Meta is weighing as it considers face recognition for a mass-market consumer device. It also shows how thin the line has grown between the surveillance technology sold to law enforcement and the military and the consumer products sold to everyone else. Increasingly, the same companies, and the same underlying algorithms, serve both. 这份许可协议是 Meta 与 Rank One 之间存在商业关系的首次已知证据，它让我们得以一窥 Meta 在考虑将人脸识别技术应用于大众消费设备时所权衡的技术类型。这也表明，向执法部门和军方出售的监控技术与向大众出售的消费产品之间的界限已经变得越来越模糊。越来越多的情况是，同一家公司和同一套底层算法同时服务于这两者。
The license Meta acquired authorizes use of Rank One’s face recognition along with its liveness detection, which checks whether a camera is seeing a real person rather than a photo or mask. It supports up to 10 million facial templates and remains active. Code reviewed by WIRED shows that remnants of Rank One’s integration—the routines that load its license and initialize its software—remained in a version of Meta’s app that shipped this month, dormant, to millions of consumers, alongside the company’s own face-recognition system. Meta 获得的许可授权其使用 Rank One 的人脸识别技术及其“活体检测”功能，后者用于检查摄像头捕捉到的是真人，而非照片或面具。该系统支持多达 1000 万个面部模板，且目前仍处于激活状态。《连线》审查的代码显示，Rank One 集成的残留部分（即加载其许可并初始化其软件的程序）仍存在于本月向数百万消费者发布的 Meta 应用程序版本中，并与 Meta 自有的人脸识别系统一同处于休眠状态。
None of the face-recognition systems tied to Meta’s smart glasses were ever active for users. Meta deleted them from the app entirely on June 5, a day after WIRED revealed that the company had quietly built an unreleased face-recognition system, internally called NameTag, into the Meta AI app—the companion software for its smart glasses, downloaded to more than 50 million phones. The system was dormant and could not be accessed by users. 与 Meta 智能眼镜相关的所有人脸识别系统从未对用户激活过。6 月 5 日，即《连线》披露 Meta 已悄悄在 Meta AI 应用程序中内置了一款名为“NameTag”的未发布人脸识别系统后，Meta 将其从应用中彻底删除。Meta AI 是其智能眼镜的配套软件，下载量已超过 5000 万次。该系统当时处于休眠状态，用户无法访问。
Meta would say almost nothing about the arrangement, declining to answer WIRED’s questions about its relationship with Rank One. Meta would not say why it licensed the software, when the relationship began, or whether it is ongoing. Rank One declined to comment for this story. Meta 对此安排几乎只字未提，拒绝回答《连线》关于其与 Rank One 关系的提问。Meta 没有说明为何授权使用该软件、合作何时开始，以及合作是否仍在继续。Rank One 也拒绝就本文置评。
Rank One Computing was founded in 2015 by a group of engineers who had built face-recognition systems at the nonprofit research institute Noblis—work that included evaluating algorithms for a US intelligence research agency. The company went public on the Nasdaq in February. Rank One Computing 由一群工程师于 2015 年创立，他们曾在非营利研究机构 Noblis 构建人脸识别系统，其工作包括为美国情报研究机构评估算法。该公司于今年 2 月在纳斯达克上市。
Rank One’s leadership is drawn from the senior ranks of law enforcement and intelligence. Its chief executive, B. Scott Swann, previously ran the FBI division that operates the bureau’s biometric databases. Its board includes a former CIA deputy director for science and technology, a former head of the FBI’s science and technology branch, and a former Pentagon official who stood up a multibillion-dollar special-capabilities office. Rank One 的领导层来自执法和情报部门的高层。其首席执行官 B. Scott Swann 此前曾负责运营联邦调查局（FBI）生物识别数据库的部门。其董事会成员包括一名前中央情报局（CIA）科学技术副局长、一名前 FBI 科学技术部门负责人，以及一名前五角大楼官员，后者曾建立了一个价值数十亿美元的特殊能力办公室。
“There’s a long history of military technologies becoming consumer products,” says Joseph Jerome, a former Meta Reality Labs policy official. “That’s arguably the story of the internet.” “军事技术转化为消费产品有着悠久的历史，”前 Meta Reality Labs 政策官员 Joseph Jerome 表示，“这可以说是互联网发展史的缩影。”
Rank One’s technology is already running in some notable places. The US Marshals Service has used a biometric identification kit built on Rank One’s technology since 2021. In West Virginia, dozens of schools have used the software to screen faces at their entrances against the state’s sex-offender registry, the company’s CEO said in 2024. Its algorithm is also bundled inside products from DataWorks Plus, a South Carolina firm, and inside LexisNexis’s Lumen platform, which lets police officers run face searches against state and regional image galleries and the FBI’s national investigative database. Rank One 的技术已经在一些重要领域投入使用。自 2021 年以来，美国法警局一直使用基于 Rank One 技术的生物识别套件。该公司首席执行官在 2024 年表示，在西弗吉尼亚州，数十所学校使用该软件在入口处进行人脸筛查，并与该州的性犯罪者登记册进行比对。其算法还被捆绑在南卡罗来纳州 DataWorks Plus 公司的产品中，以及 LexisNexis 的 Lumen 平台中，该平台允许警察针对州和地区图像库以及 FBI 的国家调查数据库进行人脸搜索。
Like other face-recognition systems, Rank One’s does not perform equally across demographic groups. In testing by the National Institute of Standards and Technology, a version of the company’s algorithm produced false matches at sharply different rates depending on a person’s sex and country of birth, which NIST uses as a proxy for race. Error rates were lowest for people born in Eastern Europe and tended to run higher for women than for men. 与其他所有人脸识别系统一样，Rank One 的系统在不同人口群体中的表现并不均衡。在国家标准与技术研究院（NIST）的测试中，该算法的一个版本在产生错误匹配时的比率差异巨大，这取决于个人的性别和出生国（NIST 将其作为种族的代理指标）。出生于东欧的人群错误率最低，而女性的错误率往往高于男性。
There are few national rules governing face recognition in the US. Many states require police to get a warrant before accessing such data, and more are folding biometric protections into their general consumer-privacy laws each year, says Eric Null, director of the Privacy and Data Project at the Center for Democracy and Technology. “But consumer-facing companies clearly crave access to high-powered facial recognition technology,” he says. “And without proper checks, the risks of this tech becoming a common consumer product are significant and largely unbounded.” 民主与技术中心（Center for Democracy and Technology）隐私与数据项目主任 Eric Null 表示，美国几乎没有管理面部识别的国家级法规。许多州要求警察在访问此类数据前必须获得搜查令，且每年有越来越多的州将生物识别保护纳入其一般消费者隐私法中。“但面向消费者的公司显然渴望获得高性能的人脸识别技术，”他说，“如果没有适当的制衡，这项技术成为普通消费产品的风险将是巨大的，且在很大程度上是无法控制的。”