Meta 推出的全新 AI 助理应用程式 Muse，旨在协助使用者自动化日常任务，包括网路购物、订位以及整合电子邮件和银行帐户。作者在测试其网页浏览虚拟机时，发现 Muse 能够顺畅地在餐厅网站选购餐点，并在 Facebook Marketplace 上高效率寻找二手家具与追踪后续进度，展现出远胜以往同类工具的浏览与执行能力。

然而，随著使用时间增加，作者发现 Muse 不断主动要求串接更多敏感资讯，包含银行余额、全量收件匣、个人证件与饮食照片，且长期记忆功能目前无法完全停用。此外，使用者预设会被纳入模型训练数据库中，这引发了隐私权倡议组织的强烈批评，认为 Meta 依然沿用操弄性的预设机制，实质上是借由聊天窗口直接收集巨量的用户隐私资料。

除了隐私外泄风险，过度将日常消费与决策外包给 AI 代理亦引发道德与认知层面的担忧。研究学者警告，深度个人化可能诱使用户透露更多隐私，甚至导致决策能力退化并丧失自主探索与培养品味的乐趣；同时代理人的推荐也可能暗藏未公开的商业导向，最终促使作者在体验后决定彻底删除该应用程式。

Meta's new AI assistant app, Muse, is designed to automate everyday tasks such as online shopping, booking reservations, and connecting to users' emails and bank accounts. While testing its web-browsing virtual machine, the author found that Muse smoothly ordered food from a local bakery website and efficiently sourced second-hand furniture on Facebook Marketplace with active follow-ups, demonstrating browsing capabilities far superior to previous tools.

However, with continued use, the author discovered that Muse constantly prompted for access to sensitive personal information, including bank balances, full email inboxes, identification documents, and meal photos, while offering no toggle to disable its long-term memory feature entirely. Furthermore, users are automatically opted in to AI training data collection, prompting severe criticism from privacy advocates who argue that Meta continues its manipulative data-harvesting practices.

Beyond data privacy risks, outsourcing everyday purchases and personal decisions to an autonomous agent raises cognitive and ethical concerns. Experts warn that deep personalization may coax users into revealing more private data while causing cognitive degradation and eroding personal taste and agency, alongside potential commercial biases in recommendations, ultimately prompting the author to delete the application.