谷歌已实质性逆转其于2023年作出的关于Gemini Workspace集成中“个人内容不会被人类审查员看到”的隐私承诺 1。这一政策变更依据更新后的Google帮助中心文章(编号16836988)以及2023年相关博客文章的10月16日编辑版本得到确认 1。新政策明确适用于受标准Gemini应用隐私声明约束的消费者Google账户(即个人@gmail.com账号),但不涵盖Google Workspace商业账户 1。
技术测试表明,在启用Connected Apps功能的消费者账户中,数据摄取具有概率性特征,且无需用户显式使用@Drive或@Gmail标签,背景对话即可自动触发对Gmail邮件和Drive文件的检索 1。检索结果主要分为四种情形:零检索、仅获取元数据或通用摘要、部分未脱敏提取(包含个人身份信息PII)、以及完整文件摄取 1。谷歌现有的正则表达式和命名实体识别(NER)自动清洗工具不足以有效去识别化包含账号、确认号等叙事数据的文本,这意味着敏感信息可能被路由至人工审核管道,进而被人类承包商查看 1。
Google has quietly reversed a 2023 commitment stating that personal content within its Gemini Workspace integration would not be viewed by human reviewers 1. This policy shift was confirmed through an updated Google Help Answer (16836988) and an October 16 edit of the original 2023 blog post, indicating that the change occurred as of October 16, 2024 1. The new terms apply specifically to consumer Google accounts (personal @gmail.com addresses) governed by standard Gemini app privacy statements, excluding Google Workspace business accounts 1.
Technical testing reveals that data ingestion into the review pipeline is probabilistic and can trigger automatically during background conversations without requiring users to explicitly tag @Drive or @Gmail 1. Once triggered, four potential retrieval outcomes may occur: no data retrieval, metadata or generic summaries, partial extraction containing unredacted personally identifiable information (PII), or full file ingestion 1. Concerns have been raised regarding the efficacy of automated de-identification measures, as Google’s regular expression-based and Named Entity Recognition (NER) cleaners appear insufficient to fully redact narrative data such as account numbers and confirmation codes before files are routed to human annotators 1.
评论
还没有评论,欢迎留下第一条。