I Thought I’d Hate AI in Home Security. I Couldn’t Have Been More Wrong - CNET

I've done years of hands-on smart home work. I was skeptical, thinking AI integration would be a hassle and a hindrance, but it turned out to be exactly what I wanted to see.

My phone buzzes with a notification, catching me off-guard while I'm out in the backyard: Person detected, package detected. It's my Google Nest Doorbell letting me know, via my iPhone's lock screen, that someone's dropped off a delivery (pet food and breakfast bars) at my front door.

A minute later, I get another alert: Package no longer seen. Uh-oh. A porch pirate already? That's been a problem in my neighborhood lately. I pop open the app to check the video doorbell's live view, but Google Home is already saving me the worry. It chimes with a reassuring message — Doorbell, front door — and the app shows a couple of friends who stopped by earlier than expected, waving at my doorbell camera and holding the package. I can hear them laughing through the two-way audio. Nothing to worry about.

The Nest doorbell isn't just a passive window onto my porch. It knew what was there, and how to describe it, thanks to AI-powered object recognition. This is a far cry from headlines about generative AI enabling cheats and exploitation — or worse, mass layoffs.

In over 100 hours of testing AI-enabled home security features, and after years of personal use, I've often found these features to be a secret sauce that's finally giving home security the edge it needs to fix long-standing problems like overwhelming alerts, confusing app management and notorious false alarms. Some worries about privacy linger, but the more I work with object detection and smart alerts, the more I feel like we're getting something right.

Well before last month's introduction of a chatty Gemini AI that saw and discussed multiple objects as someone moved around an office, Google was quietly training Nest algorithms to recognize the difference between a package and a person. Now you can find object detection and recognition on nearly every smart home cam, from Arlo's 2K-resolution devices and Ring's plentiful doorbells to Eufy, Lorex, Reolink, SimpliSafe and many more. We've opened our doors to a quiet AI revolution where people, amazingly, have few complaints.

I've been working in smart home tech for a decade now, and these days I review, among other things, AI-equipped products in my role as CNET's home security editor. (When that notification buzzed, I was in the middle of setting up a backyard camera for my latest review.)

At my townhome in Bend, Oregon, the security setup can shift suddenly depending on what I'm testing — anything from a new SimpliSafe indoor sensor kit to the latest lever-based lock from Schlage — but I keep a few core items in play as my personal devices. That includes a video doorbell, a deadbolt smart lock and a backyard cam, plus a couple of smart displays for voice commands or other controls.

That setup is easy (each device takes around 30 to 60 minutes to install), works with Alexa or Google Home, and is simple to teach to family and friends. It also comes packed with a collection of the latest algorithms to detect and filter recognized objects.

This tech is more affordable than it's ever been. The cameras themselves are available well below $50 for those who want to save. Higher-end video doorbells are around $100 to $200, much lower than several years ago, while home security kits can start at a few hundred dollars. AI detection is either completely free or available as an add-on to services that charge $3 to $8 per month. This means we can recommend technology like this to people who wouldn't have been able to afford it in years past.

Smart detection isn't just making our homes safer, it's giving our busy minds a break.

"The smart security segment stands out for its consistent innovation and strong value proposition," says Adam Wright, research manager and smart home specialist at market intelligence firm IDC. "AI-enabled cameras and video doorbells, in particular, have driven the continued growth and interest in smart security devices, thanks in part to a clear value proposition of safety, security and enhanced capabilities."

While no devices are perfect, I do see potential: This technology is practical and easily customizable, with real-world benefits for our homes and families. I'm not dismissing potential issues with privacy or how companies manage customer data, but this home tech uses today's AI training models with a light touch. If we must get used to artificial intelligence appearing everywhere, these applications show that it doesn't have to be a bad thing.

We're only now at the beginning of the generative AI revolution kicked off by ChatGPT in late 2022, which itself builds on several decades of behind-the-scenes developments across the broader landscape of artificial intelligence. The hype for gen AI has been nothing short of spectacular, but some people see a misplaced emphasis, something I've been calling the "laundry and dishes" effect.

"You know what the biggest problem with pushing all-things-AI is? Wrong direction," the fantasy and sci-fi author Joanna Maciejewska wrote in a March tweet that quickly went viral. "I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes."

Home security is currently winning that AI challenge. I don't have a robot butler in the laundry room (it's more of a laundry closet, really), but I do have an app that alerts me when a family member walks up the driveway and that doesn't bother me with reports of every vehicle or jogger that passes the front yard — unless I want it to.

In many ways, home security algorithms are like the LLMs (large language models) used in ChatGPT, Gemini, Copilot and other popular online AI. LLMs are made to take apart the details of language, objects or logic into data sets they can learn, then put it all back together to answer questions or create useful info. Over time and with careful training (along with massive amounts of data), these models grow more accurate, like when you take a test over and over, until they can reliably distinguish people from cars or pets, and even recognize individual faces. The devil in the details is how they're used.

A few taps on a phone or tablet touchscreen, like from the Blink camera app, let me filter objects to get only notifications about animals in the backyard. Or I can block out all patio motion detection with a privacy screen, or adjust motion sensitivity if it's triggered by every jogger on the sidewalk. With a bit of settings work, I can receive only the information I want about what's happening in and around my home.

AI filters make sure that security cameras only bother us when important objects are detected at home.

First-time users are often surprised by how far it's all come. My friend Carl, using the Arlo app for his first video doorbell, was pleased by how well it paired with existing smart devices. He also noted how "crisp" the visuals were, another field that's come a long way in recent years and enables that AI accuracy.

Such targeted setup and control is what security system brand SimpliSafe calls "proactive security," and it had previously been limited to high-end commercial or "elite residential" applications. But now AI is bringing object recognition to our pockets — and you don't even need a residential single-family house to use it. Plenty of indoor security cams, sensors and even peephole doorbells are made for renters of all kinds.

Object recognition is only the beginning of the chores that AI-enabled home security tech can perform. Another of my friends gets AI alerts from Furbo while he's away at work. The pet cam reports when his young dog has jumped on the couch, when he's chewing on something or when it looks like he vomited (occasionally, it also thinks his robot vacuum is the dog). It's a far cry from AI tools that mistakenly recommend, say, putting glue on pizza.

Another acquaintance customizes motion detection to get alerts about when her kid wanders out of the bedroom after naptime while she's working in a different room. Sound recognition, meanwhile, is telling us all when it hears a siren, when an alarm goes off or, as with Ring's Alarm kit, when it detects glass breaking up to 25 feet away. You don't have to be home to receive any of these notifications, but you can take action, like calling in through a smart speaker or contacting emergency responders, once again for free or a few bucks a month.

Home security software isn't a full-fledged nanny or doorman, but it's getting closer with every update. Oh, and it's helping us stay calmer, too.

Let's talk about anxiety. All that tech can make people nervous, whether from sheer information overload or the paranoia of constant security alerts and false alarms. When our phones are gateways to every hiccup a home system may have, it's easy to grow jumpy.

Privacy fears have been an issue since the smart home became a thing. A 2023 study from construction company CraftJack showed that 1 in 4 users worry about their cameras being hacked, and over 60% believe "their devices are always listening." (They aren't, and hacking home security remains more or less unheard of.) And even without concerns about strangers peering in, security systems can still provide a torrent of safety information that can be a lot to process.

The key here is to have well-made software filters combined with object identification, which together act like a club bouncer, letting only the most legitimate and best-dressed notifications through to our harried minds. As I noted above, the Blink Outdoor cam that watches my backyard avoids mentioning anything but a person unless I specifically turn on the animal settings to watch for a pet.

Today's doorbells give immediate alerts for packages, It's useful and generally stress-free.

Before smart alerts, motion detection would shove every possible notification in front of our eyes, and without algorithmic recognition it was a lot easier to trigger cameras or sensors with birds, swaying tree branches and passing cars. Add a floodlight that can trip on and off throughout a night of false alarms, and brains truly get fried. Even the least suspicious homeowner could start jumping at shadows.

With the advent of security AI, cameras and lights have begun passing along info to you only when it's pertinent, like alerts about a nearby suspicious human. Animals, cars and distant passersby get ignored. Equip cams with motion detection zones that let you draw a circle around the areas for AI to analyze, and users can ensure cameras stop tripping on active sidewalks or nearby paths.

Other studies have shown that even as some home security can cause "surveillance-related stress," what researchers call "ethical" system designs with privacy safeguards can reduce fear, as well instill a sense of security. My buddy Jake, who uses multiple cams from different brands outside his house, told me that, "It especially makes me feel good to know that nobody's entered the house If I come home to find a door was left unlocked or a window open."

This common sentiment is also why the American Institute of Health Care Professionals has chimed in with suggestions on how smart home devices can help tackle anxiety, like cameras listening for smoke alarms or smart locks that can secure the door behind you every time. Companies like Lorex, known for its subscription-free home cameras, and insurer Nationwide have begun underlining how today's home security devices can save on stress and anxiety.

And security companies are striving to focus systems even more. Reolink and SimpliSafe, for example, are working with more advanced algorithms that use the model training I mentioned above to learn the movements associated with strangers loitering, skulking or generally acting with ill intent — as opposed to the DoorDash driver trying to get your curry over while it's still warm.

From AI saving us false alarms to new security apps that let us ask communities questions, home security is starting to fix its anxiety problem.

If smart, filtered alerts still leave you nervous, the latest software offers a further option for peace of mind: Turn to your community for crowdsourced answers, like Ring's Neighbors app or similar options from Reolink, Eufy and Arlo. It's surprisingly effective.

"These four came down our very long driveway at midnight," a worried neighbor posted on the Ring app last month. "High school kids in formal wear. Last night was prom," A local member quickly responds, deescalating the problem. Another example: "Our car was hit today and the driver drove away. I need Ring footage of the incident if anyone has it," a neighbor wrote. "Did someone nearby get my package? It shows as delivered but I've gotten no mail today or yesterday," asked a third.

With a bit of help from AI camera detection, these users get the answers they seek or useful contact info. Strict moderation and regional filters separate the apps from the doomposting you can run into on NextDoor or Facebook.

But there's also a catch: AI systems need data, the more the better, to be effective. In this case, that's data about you, your friends and neighbors, and even how people move around your house. Fortunately, that's largely opt-in. From Brinks to Arlo's Caught in the Act community, these networks need your permission to use your uploaded videos for their AI training models — or onto broader social networks like Facebook (Eufy, in particular, suggests users visit the Facebook Group for owners) where data scraping is both common and legal.

Philip Kolterman, senior vice president of digital transformation and IT at security system creator Brinks Home Security, says participation is optional for the Brinks program and users must nominate a specific camera if they agree. "Video may then be seen by engineers working to improve the tooling," he says, but third parties can't get their hands on it.

Other services don't give you choices about what cameras to monitor, or what they'll do with data when you sign up for smart AI alerts. That's ringing alarm bells for some users, so let's address an issue marring an otherwise sunny view.

My Nest Doorbell, with its years of AI model training, is extremely accurate. I can't think of one time that Google's algorithm has gotten a package or human mixed up — unlike some object detection, which occasionally skips packages to focus on the people putting them down. Or my colleague's Eufy cam, which has been known to identify her cat as a human.

Nest wants to push the envelope even further with a newer feature that I described to a gathering of friends as "like making a phone contact profile, but for your face." Unimpressed, they collectively declined.

Nest sees this feature as enabling family-first routines. When I contacted Julie Zhu, product manager for Google Nest, she said users could "build automations related to a specific person, like personalized doorbell chimes." But that can sound like a tough trade. "It seems like something I would have liked 10 years ago," said another friend of mine, who simply doesn't hold much trust in tech companies these days.

Nest's "familiar face" detection, available with a subscription, is like services offered by Eufy, SimpliSafe and a growing number of other security brands. AI is now more than advanced enough to recognize faces based on profiles you save from contacts and photos, telling you if the person at the door is a welcome acquaintance or a stranger. For its beta program, SimpliSafe suggests that you could even save photos of your dogwalker's face so you know when they're waiting at the door.

Does that make your privacy senses tingle? You're not alone: I couldn't generate much interest in getting "profiled" from those I talked to, and I can't imagine dogwalkers being thrilled about having their faces recorded. It feels like a privacy risk to trust your personal features to the cloud, which can be subject to security breaches or used for who-knows-what kind of AI training.

And if you choose cloud video storage or similar services, then even uglier risks arise, like the ability for errors to let strangers see through your home cameras or even allow company employees to spy. We try to recommend security cameras with local storage and no subscription element so you can keep as much data offline as possible, but the threat of data theft remains, well, in your face.

This sticky spot in an otherwise smooth ride is giving security companies pause, too. Some are focused on end-to-end encryption and client-side data storage to help keep face details off the internet. That's similar to how Apple protects Face ID profiles — and what it's doing in its new Apple Intelligence services — except you're offering up data about your acquaintances, not just yourself.

The law also has a bone to pick with face recognition. States have started considering legislation to limit facial recognition, particularly in public places. Illinois' Biometric Information Privacy Act goes a step further and limits the use of facial recognition by private companies. That's why you can visit the small print on Google's Nest Aware plans and see that its familiar face technology is blocked entirely in Illinois. I won't be surprised when other states pass similar laws.

That raises questions about the cops, too: Can police access your face profile if they want to? Law enforcement can request home security footage from a business's cloud with a warrant or in case of a life-or-death emergency (although companies like Ring have tightened up their rules). We don't know if that pertains to new technology like facial data or how cops could use it.

Other brands bypass this issue by investing in an alternative recognition option: your voice. Reolink, Anker and SimpliSafe have varying levels of voice recognition technology that map out and identify specific voices. Eufy, in particular, is optimistic about using parent company Anker's VoicePrint technology to manage access control in video doorbells and smart locks. That could lead to accurate person identification without the need to sacrifice facial data and the potential issue that carries.

But for some, consigning their voice to the AI netherworld also sounds suspect. Scarlett Johansson certainly isn't a fan of the possibilities, and we everyday home-havers must be careful about who owns our voices, too. At this point, these deeper recognition features are still locked behind paywalls, and paying for an AI close-up isn't as tempting as some brands seem to think. There's a reason Alexa lets you turn off Voice ID if you don't want it.

While home security AI offers an example of healthier artificial intelligence, it's not a panacea — more like a diet plan. Practices like potentially invasive face and voice profiles underline the universal struggles we're encountering. Home security hasn't solved those problems, but it's giving us a real-world zone to hash them out.

With the right settings, outdoor cams can do a whole lot... without the need to scan faces.

The last time I mowed my lawn with my AirPods on (I'm a huge fan of the third gen), Siri happily chirped up and passed along a Nest message, "Animal detected, zone one, front door doorbell, Person detected, zone one, front door doorbell." My neighbor had gone chasing after her escaped shih tzu again. I stopped mowing, just in case: That tiny fluffball has horrible situational awareness. It's rare to see all my smart home platforms so happily working as a team.

My Nest Doorbell doesn't greet people like a concierge when they get home — although I'd bet that's coming, probably with a new wave of celebrity voice mimicking. But it does give me peace of mind, and those "package detected" alerts never lose their usefulness. I've yet to convince anyone to let me create a doorbell face profile, but I think more uses will emerge, like family members who can get special permissions for smart locks. That's one of the strengths of home AI: It has time on its side, and a lot to like when you get used to it.

This crossroads of AI integration and happy people is a sign that not all AI endeavors lead straight to dystopia. Can home security serve as a roadmap for incorporating AI tech into our daily lives without invoking a Black Mirror episode? I hope so. It's helping people move past paranoia about strangers, which I'd call a good start.

I won't dare to predict precisely what home security will become in the next several years. Face and voice recognition are likely here to stay, although they'll need to find long-term value instead of mere tech dazzle. And I wouldn't bet against Google's latest Gemini Agents and Gems — build-a-bot options for apps or personalized AI on our devices — growing more integrated, answering complex smart home questions and making security assessments on the fly.

I also wouldn't count Arlo out, either. New improvements to the company's Secure plan have added the ability to recognize specific people and vehicles, helping fine-tune the way homeowners can engage with their home security systems as they add more familiar faces to their daily lives.

SimpliSafe's AI-empowered agents, even in beta, could be a blueprint for the future of professional home monitoring, too, albeit for those willing to pay around $30 per month. Thankfully, prices continue to fall for both advanced software and monitoring subscriptions.

If this AI home intrusion does herald the end of the world as we know it then, as the band says, I feel fine. Just keep those data breaches at bay, and I can choose exactly which AI features I participate in and how much they affect my daily life. The balance of privacy (with a few question marks), utility and control is a solid mix. Whether it will continue to stay balanced is unclear, but for now, our AI security guards have made themselves at home.

Visual Designer | Zooey Liao

Video | Mike Viney

Senior Project Manager | Danielle Ramirez

Project Manager | Minphy Liao

Director of Content | Jonathan Skillings

Editor | Corinne Reichert