个人人工智慧助理的实用性取决于被授予的数据与权限，这使科技公司在设计硬体时面临隐私与功能之间的张力。去年，Amazon 收购了开发穿戴式笔记设备的人工智慧新创公司 Bee 后，修改了该产品预设在背景被动录音并生成对话摘要的行为，要求使用者必须先按下按钮才能开始撷取音讯。此外，尽管 Meta 等公司尝试透过移除相机镜头来提供妥协方案，但无相机设备并未解决全天候聆听的隐私疑虑，例如 Apple 即将推出、尚未开放消费者试用的全天候聆听 Apple Watch 功能，就因可能在未提示取得他人同意的情况下拾取附近对话而面临审查。

这种关于权限的讨论正延伸至新兴的人工智慧代理应用程式，例如 Meta 的 Muse 或 OpenAI 的 dots。这些工具允许使用者连接个人应用程式甚至银行帐户，以便在背景主动运行并代为执行操作。由于这些工具具备全天候运行与自主运作的特性，决定授予人工智慧多少存取权限变得更加关键。

然而，统计数据与市场趋势显示，消费者与金融机构对高度自主的人工智慧仍抱持极高戒心。根据 Ping Identity 本月初发布的一份报告，在全球受访的 11,000 名消费者中，仅有 3% 的比例（即 97% 持保留或反对态度）对于人工智慧在未经其最终批准的情况下执行例行操作或进行交易感到放心。这一明确的统计趋势也解释了为何华尔街最大的几家银行目前正积极封锁人工智慧代理取得客户帐户资讯的控制权。

The usefulness of personal AI assistants depends on the data and permissions granted to them, creating a tension for tech companies designing hardware between privacy and functionality. Last year, after Amazon acquired Bee, an AI startup that made a wearable note-taking device, it altered the product's default behavior of passively recording and generating conversation summaries in the background, requiring users to press a button first to capture audio. Furthermore, while companies like Meta have attempted a compromise by removing the camera lens, camera-free devices do not resolve privacy concerns around always-listening AI; for instance, Apple is already facing scrutiny over forthcoming, not-yet-available always-listening Apple Watch features that can pick up nearby conversations without prompting for others' consent.

This discussion regarding permissions is extending to emerging AI agent applications, such as Meta's Muse or OpenAI's dots. These tools allow users to connect their personal applications and even bank accounts so they can work proactively in the background and take actions on their behalf. Because of the always-on and autonomous nature of these tools, deciding how much access to grant AI becomes even more consequential.

However, statistical data and market trends indicate that consumers and financial institutions remain highly cautious about highly autonomous AI. According to a report released earlier this month by Ping Identity, out of 11,000 surveyed consumers worldwide, a mere 3% ratio (meaning 97% are hesitant or opposed) felt comfortable with AI taking routine actions or making transactions without their final approval. This clear statistical trend explains why the biggest banks on Wall Street are currently actively blocking AI agents from gaining control of customer-account information.