软体分析显示，Meta 已将一套未发布的脸部辨识系统（内部称为「NameTag」）悄悄整合至其 Meta AI 伴侣应用程式中，该程式已被下载超过 5,000 万次（50 million times）。尽管 Meta 公开声称正采取周详的方法且仍在探索该技术，但包括侦测、裁剪和将脸部编码为生物特征签章之三款 AI 模型在内的核心组件，早在 1 月就已部署至用户装置。该系统旨在将 Meta 智慧眼镜所捕捉的脸部特征，与储存在用户手机本地的资料库进行比对，而该资料库设定为接收来自 Meta 伺服器的更新。

此一发展引发了重大的隐私疑虑，使人联想到 Meta 过去在脸部辨识领域所引发的争议。于 2021 年，Meta 在面临严格审查后，关闭了类似的照片标记系统并删除了超过 10 亿个（1 billion）脸部特征。这段历史包括在 Illinois 支付的 6.5 亿美元（$650 million）集体诉讼和解金、在 Texas 达成的 14 亿美元（$1.4 billion）和解协议，以及 2019 年向 Federal Trade Commission 和 Department of Justice 支付的 50 亿美元（$5 billion）和解金。隐私倡导者（包括要求 Meta 废除 NameTag 的 70 多个倡导团体）警告，该技术可能会将用户的智慧眼镜转化为分布式监视网络。

尽管存在这些疑虑，且 Meta 公开声称并未建立集中式的脸部资料库，但 2 月的内部文件显示，该公司曾计划在政治环境繁忙的窗口推出此功能，以减少批评。虽然 Meta 继续将 NameTag 描述为仅是探索，但安全研究人员（如 Cooper Quintin 与 Buchodi）已证实其底层代码已近乎就绪，Buchodi 甚至成功使用预载的 Michel Foucault 脸部特征对该系统进行了测试。Meta 此前也曾计划将此技术引进辅助使用场景，例如协助视障人士识别熟人。

A software analysis reveals that Meta has quietly integrated an unreleased facial recognition system, internally named "NameTag," into its Meta AI companion app, which has been downloaded over 50 million times. Although Meta publicly claimed to be taking a thoughtful approach and still exploring the technology, core components including three AI models—designed to detect, crop, and encode faces into biometric signatures—were deployed to user devices as early as January. The system is designed to match faceprints captured by Meta's smart glasses against databases stored locally on users' phones, which are configured to receive updates from Meta's servers.

This development raises significant privacy concerns, recalling Meta's controversial past with facial recognition. In 2021, Meta shut down a similar photo-tagging system and deleted over 1 billion faceprints following intense scrutiny. This history includes a $650 million class-action settlement in Illinois, a $1.4 billion settlement in Texas, and a $5 billion settlement with the Federal Trade Commission and Department of Justice in 2019. Privacy advocates, including more than 70 advocacy groups that demanded Meta scrap NameTag, warn that the technology could turn users' smart glasses into a distributed surveillance network.

Despite these concerns and Meta's public assertions that it is not building a centralized face database, internal documents from February show plans to roll out the feature during a busy political window to minimize criticism. While Meta continues to describe NameTag as a mere exploration, security researchers like Cooper Quintin and Buchodi have verified that the underlying code is nearly functional, with Buchodi successfully testing the system using a preloaded faceprint of Michel Foucault. Meta had also previously planned to introduce the technology under assistive use cases, such as helping blind individuals identify acquaintances. (Key numbers: 2)