Smart assistants: trust and privacy debate

The smart-home fight used to be about convenience. Now it is about trust. This week’s coverage tied together two changes that make that plain. Google is pushing Gemini into the home as a more natural way to control lights, thermostats, speakers, cameras, and routines. Amazon is doing the same with Alexa+. But the more these assistants sound like people, the more the old questions harden into practical ones: what can they control, what data do they need, and who in the house will actually believe them enough to use them (blog.google, amazon.com). Google’s pitch is easy to understand because it fixes a real annoyance. For years, smart homes worked best when you spoke in clipped commands. Gemini for Home is supposed to handle ordinary language instead. Google says Gemini now replaces Google Assistant on Nest speakers and displays in early access, and it can interpret more conversational requests in the Home app and on phones. The company also says the Gemini mobile app can control the same devices your Google Home account can access, including lights, climate devices, blinds, TVs, and appliances added later (blog.google, support.google.com, support.google.com). That sounds like a simple upgrade until you reach the front door. Google draws a bright line around security-critical actions. Its own help page says Gemini home controls are “for convenience only” and warns users not to rely on them for safety or security tasks. It explicitly says Gemini cannot unlock a door for you. If you ask for an unsupported security action, the app hands you back to Google Home instead of doing it directly (support.google.com). That is not a bug. It is a design choice, and a revealing one. Amazon made a different choice long ago, and it still shows. Alexa supports voice unlocking for compatible smart locks through Key by Amazon, but only after the user turns on “Unlock by Voice” and sets a four-digit code that must be spoken to complete the action (amazon.com). In one narrow sense, that is more capable than Google’s current Gemini setup. In another, it exposes the tradeoff with unusual clarity. The assistant that can do more at the threshold of your home also has to hear more, process more, and be trusted more. That matters because Alexa’s privacy story got worse, not better, as Amazon moved deeper into generative AI. Amazon’s current privacy documentation says that when you speak to Alexa, a recording of your request is sent to Amazon’s cloud, where it is processed and turned into a transcript (amazon.com). In 2025, Amazon also removed the Echo setting that had allowed some users to keep voice recordings from being sent to the company for processing, a change widely reported at the time as part of the shift toward AI features that depend on cloud compute (techcrunch.com, pcmag.com). Amazon still offers deletion controls in the Alexa app, but that is not the same thing as not sending the audio in the first place (amazon.com, amazon.com). And that brings the story to the people who are usually treated as an afterthought in consumer AI launches. A Georgia Tech-led study released on March 31 found that older adults who live alone and already use voice assistants still do not automatically trust AI advice. The researchers interviewed 23 older adults and found that clear explanations were essential. Not flashy confidence scores. Not a smoother voice. Explanations that made sense in plain language and matched the situation. Some participants said they already felt excluded by products designed as if everyone wants the same style of interaction (news.research.gatech.edu, cc.gatech.edu). That is the part the platform comparison can miss. The hard problem is not whether Gemini sounds more natural or whether Alexa can unlock a smart lock with a PIN. The hard problem is whether a household can form a stable mental model of what the assistant is doing. Google’s refusal to let Gemini unlock doors may frustrate power users, but it is legible. Amazon’s broader lock support is useful, but it asks for a deeper reserve of trust at the exact point where a mistake matters most. For a family choosing one system for parents, kids, and older relatives, the decisive feature may be the one that explains itself before it acts (support.google.com, amazon.com, news.research.gatech.edu).