The privacy discussion around smart home devices tends to get stuck on the most dramatic scenarios — leaked camera feeds, unauthorized recordings, the occasional news story about a hacked baby monitor. These incidents are real, but they're not the primary privacy issue. The primary issue is the systematic, continuous, intended data collection that happens by design, within the terms of service you agreed to.
Understanding what's actually being collected, by whom, and what it's used for is the starting point for making informed decisions about what to invite into your home.
Smart Speakers: The Always-On Microphone
Amazon Echo, Google Nest, and Apple HomePod devices use wake word detection that runs locally on the device — they're not streaming continuous audio to the cloud. This is technically accurate and commonly misunderstood. The device listens locally for the wake word; once triggered, it sends the subsequent audio to the cloud for processing.
The data collection issues are subtler than continuous listening:
- False positive recordings — wake word detection is imperfect. Phrases that sound like "Alexa" or "Hey Google" trigger recording. Amazon disclosed in 2019 that human reviewers listen to a sample of Alexa recordings, including accidental ones, for training purposes.
- Query logs — everything you ask is logged with timestamps and associated with your account. Over months, a query log is a detailed record of your concerns, interests, health questions, and household patterns.
- Presence inference — smart speakers with motion sensors or always-on microphone analysis can infer when people are home, how many people are present, and routine patterns, even from ambient sound analysis.
- Third-party skill data — Alexa "skills" and Google "actions" are third-party apps. When you invoke a third-party skill, the interaction data flows to that third party's servers. Those parties have their own, often weaker, privacy policies.
Amazon, Google, and Apple all provide interfaces to review and delete stored voice recordings. For Amazon: Alexa app → More → Activity. For Google: myactivity.google.com. For Apple: Settings → Siri & Search → Siri & Dictation History. Deleting recordings is worth doing periodically.
Smart TVs: Automatic Content Recognition
Most smart TVs ship with a feature called ACR (Automatic Content Recognition). It works by taking periodic screenshots of whatever is on your screen, hashing or fingerprinting them, and comparing them against a database of content to identify what you're watching — including content from external inputs like a game console or Blu-ray player. The data is sent to the TV manufacturer and often to advertising partners.
ACR doesn't need your streaming service's data. It sees the pixels on your screen. Whether you're watching Netflix, a DVD, or live cable from an HDMI input, ACR knows. Vizio, LG, Samsung, and Roku have all faced regulatory scrutiny over ACR practices. Vizio paid a $2.2 million settlement to the FTC in 2017 over undisclosed ACR collection.
ACR is almost always enabled by default and the setting to disable it is buried. The usual location: Settings → Privacy → Smart TV Experience, or equivalent. The naming varies by manufacturer; the function is consistent. Disable it, and the TV's "smart" features still work — you just stop feeding viewing data to the ad ecosystem.
Video Doorbells and Cameras: Network-Level Concerns
Amazon Ring, Google Nest cameras, and similar devices store recorded footage in the cloud. This creates several distinct risks:
- Law enforcement requests — Amazon has a documented history of providing Ring footage to law enforcement in response to requests, including without user notification in some emergency cases. Ring operates a "Neighbors" platform that has formalized relationships with hundreds of police departments.
- Facial recognition — some Ring and Google Nest devices offer opt-in facial recognition. The face data associated with this feature is stored on company servers.
- Third-party data sharing — Ring's privacy policy has historically permitted sharing data with a wide range of third parties. The specific scope has changed over time in response to public criticism.
The most robust alternative for camera footage you want to control: local recording via an NVR (Network Video Recorder) with cameras that don't phone home. Systems built on open-source software like Frigate, combined with local-only cameras (reolink, Amcrest with cloud features disabled), keep footage on hardware you own. This requires more setup than plug-and-play cloud systems; it eliminates cloud exposure entirely.
Network Segmentation: Isolating Your IoT Devices
One of the most practical defenses against smart home device overreach is network segmentation: putting IoT devices on a separate network (VLAN) that can't communicate with your computers, phones, or NAS. This doesn't stop a device from sending data to the internet — it stops a compromised device from being able to pivot to your other devices on the same network.
| Approach | What It Prevents | What It Doesn't Prevent |
|---|---|---|
| IoT VLAN | Compromised device accessing your main network; lateral movement | Device calling home to manufacturer servers |
| DNS-based blocking (Pi-hole) | Connections to known tracking/ad domains at the DNS level | Connections using hardcoded IPs; encrypted DNS bypasses |
| Firewall egress rules | IoT device communicating with unexpected servers | Requires knowing expected destinations in advance; high maintenance |
| Local-only alternatives | Cloud data collection entirely | Requires more setup; may lose vendor features |
A Pi-hole running on your network intercepts DNS queries from all devices and blocks domains on its blocklists. It's a coarse but effective tool — it doesn't see encrypted traffic content, but it can block the DNS resolution for known tracking domains. Many IoT device telemetry endpoints are standard domains that appear on common blocklists.
Local-Only Alternatives: Home Assistant
Home Assistant is an open-source home automation platform designed to run locally on your own hardware — a Raspberry Pi, an old server, or a purpose-built device. It supports thousands of devices through integrations, including most major smart home brands, and can operate entirely without cloud connectivity for the devices that support local control.
The trade-off is setup complexity. A cloud-based ecosystem like Amazon's or Google's is plug-and-play because they've invested heavily in the onboarding experience. Home Assistant gives you full control and local operation, but you're responsible for configuration, updates, and troubleshooting. It's the right choice for anyone who has already moved toward self-hosting other services and is comfortable with the operational overhead.
Local control doesn't inherently mean the device hardware is trustworthy. A device that supports local control via its LAN API may still phone home when it has internet access. Firewall rules or network-level blocking are needed to enforce local-only operation on hardware you don't fully control.
Practical Starting Points
If overhauling your entire smart home setup isn't realistic right now, these actions have the highest impact per unit of effort:
- Disable ACR on your smart TV (settings are in the privacy section on most models)
- Review and delete stored voice recordings quarterly for any smart speakers you keep
- Disable third-party Alexa skills and Google Actions you don't actively use
- If your router supports VLANs, segment IoT devices onto a guest network with client isolation enabled — this at minimum stops devices from communicating with each other and your main devices
- Review companion app permissions on your phone — smart home apps routinely request location, microphone, and contacts access they don't need for basic operation
The goal isn't to eliminate smart home devices if you find them useful — it's to have an accurate picture of what they collect, make deliberate choices about what you keep, and take the available steps to reduce collection where you can.