It’s a year of refinement for the Galaxy Watch. With the new Galaxy Watch 9 and Galaxy Watch Ultra 2, which are being announced today, Samsung is more or less taking what people like about its watches and tweaking those elements to be better. The company has been doing this for the past few generations of its watch line, but this year’s updates seem focused on shoring up Samsung’s wearable weaknesses. Namely, battery life and preventive health features.

The Ultra 2, now $50 more expensive at $699.99, gets a few key spec bumps. The 47mm watch packs in a whopping 800mAh battery — a major 35 percent increase from the 590mAh battery in the original Ultra. I got around 70 hours on the original Ultra, so I’m curious to see how this battery boost translates to everyday usage. I was worried the larger battery would necessitate a chonkier watch, but surprisingly, the case is 12 percent thinner, measuring 10.7mm. The standard strap is also about 1mm thinner at 1.9mm, and a bit softer to the touch. 

Another tweak for outdoors enthusiasts: The watch now has a 5,000-nit display with thinner bezels around the sides. That level of brightness might seem excessive, but it’s meant to improve visibility in direct sunlight. I wasn’t invited to see the watches in person, but my colleague Owen Grove saw the watch indoors and said the display looked slightly brighter than the Watch 9, which has a 3,000-nit display. That gives the Ultra 2 the brightest smartwatch screen and among the biggest batteries.

The Ultra 2 is also getting a new trail run tracking feature, which seems like an attempt to compete with Garmin more so than the Apple Watch Ultra line. It allows you to download GPX files for GPS routes to enable real-time navigation and elevation guidance. You’ll also get an estimated completion time and real-time hydration alerts. One note: The algorithm essentially predicts how much you’ve sweat based on workout intensity and demographic data — the sensors themselves aren’t measuring how many electrolytes or how much water you’ve actually lost. Divers also get IP69K, 10 ATM, and EN 13319 dust- and water-resistance rating, plus a bespoke diving app from scuba equipment maker Mares. The app will arrive later this year and will monitor descent speed, record time at a given depth, plus give ascent guidance.

The Galaxy Watch 9, which is seeing a $30 price bump to $379.99, is mostly the same as last year’s Galaxy Watch 8. The most notable difference is that the Watch 9 (and the Ultra 2) ditch Samsung’s proprietary Exynos chipset for Qualcomm Snapdragon Wear Elite. Presumably, this is to help with power efficiency and AI processing. (Qualcomm announced the chip earlier this year at Mobile World Congress as an “AI-wearable first” chipset.) This could be to perhaps boost Gemini performance, and one new update with Wear OS 7 is you’ll be able to raise to talk without using the Gemini wake word. Samsung says it’s also tuned Gemini so it won’t trigger when you lift your wrist to check the time. It’s hard to gauge any performance boosts at a hands-on; some of my colleagues noticed a slight latency while scrolling through screens, but that could easily be due to demo models and wonky connectivity. We’ll have to see how the watches fare under normal testing conditions.

Battery capacity gets a slight boost on both the 40mm and 44mm watches. The smaller watch gets a roughly 20 percent increase to a 390mAh battery, while the larger one gets a more modest bump from 435mAh to 445mAh. The straps are 17 percent lighter with a new “air pocket structure,” which allows for a gap between the fabric and wrist for what Samsung promises is a more comfy, breathable fit.  

The bulk of the Galaxy Watch 9’s updates come in the form of four new preventive health metrics, which are also coming to the Ultra 2. The first is a Vitals feature. After a week of consecutive wear during sleep, the Samsung Health app will generate a baseline for blood oxygen, heart rate, heart rate variability, skin temperature, and respiratory rate. In the morning, you’ll get a proactive alert if any deviate from your baseline, alongside some tips on how to get back on track. This is basically a dupe of Apple’s Vitals feature, which was released with watchOS 11 in 2024.

It’s also adding a Heart Health Score and a Daily Cardio Load metric. The former generates a score out of 100 every day based on your eight-day sleep time average, moderate to vigorous activity level, BMI, and vascular load. The latter helps you balance training versus recovery based on workout intensity, maximum training load, and real-time heart rate data. The idea is to help you determine if you’re overtraining, undertraining, or doing just the right amount. Again, these aren’t new ideas. We’ve seen similar features from Oura, Whoop, and Fitbit (now Google Health) over the past few years.

The most unique health metric is the new Fitness Index. It’s a radar chart that graphs how your strength, flexibility, endurance, cardio, and body composition compare to other Samsung users, based on factors like your VO2 Max, heart rate, step counts, activity time, and bioimpedance analysis results. You might be most familiar with radar charts if you play roleplaying games — it doesn’t tell you whether your metrics are good or bad, but rather it’s a method for visualizing your relative strengths and weaknesses.  

This generation of Samsung watches feels like a mix of catching up to rivals’ health features and refining what already works. That’s not the worst combination, except for the fact that all the watches are more expensive. The company has been slowly increasing the prices of its wearables over time — though geopolitical factors probably play some role in this year’s bumps, too. Even so, you never like to see prices go up.

The watches are available for preorder starting today and are expected to ship starting August 7th.

I first slipped on Halliday’s original smart glasses at CES 2025. I was not a fan. The glasses had a tiny, movable display window embedded into the frame that was incredibly finicky to see, and my 30-minute demo left me with achy eyeballs. It was an interesting concept with terrible execution, especially compared to the several other smart glasses on the show floor. So I was skeptical when a few weeks ago, Halliday reached out to say its second attempt at smart glasses was much better.

Well, having tried the Halliday Gen 2, I’m happy to report these are better glasses. That secret display window is gone, replaced by more traditional but discreet waveguides. They’re not like the Meta Display’s nigh-invisible screens. You can still spot the reflection if angled just right, but they’re considerably easier to see. And inside of a single-lens display, the Gen 2 has one for each eye, giving you a much larger screen to look at. Spec-wise, the Gen 2’s displays are monochrome green, with a maximum brightness of 1,600 nits and 600 by 300 pixel resolution per eye.

On my face, the Gen 2 has a nondescript design but is very lightweight compared to other display glasses I’ve tried. The display was also easily visible inside a dark bar, though I didn’t get a chance to see how it’d fare in direct sunlight. The four-mic array was able to pick up commands despite the ambient noise, and Halliday told me that it gets about 12 hours of typical use. That’s partly because these aren’t camera glasses. The exclusion of the camera was a deliberate choice, both for privacy and to extend battery life. That’s because these are more so meant to be a tool for work meetings.

With the Gen 2, Halliday is launching a feature called Meeting Flow. It’s meant to help you follow along in a conversation, using AI to surface context or facts without needing to look at your laptop or phone. I’ve seen this in other smart glasses to varying degrees of success. (Mostly non-success.) More interesting are the so-called Thread Tracker, Decision Confirmation, and Commitment Check features. The idea is to provide real-time updates about what decisions have been made in a meeting, what items need following up on, and a visualization of a meeting’s overall conversation flow. For example, Thread Tracker might show that a team meeting started off talking about budget allocations, before moving on to potential new hires and timelines for those hires. Decision Confirmation might show that the meeting concluded that all new hires should be finished by a specific date. Lastly, Commitment Check verifies if action items have been assigned (and to whom) with a clear deadline. The glasses will also provide post-meeting transcripts and summaries.

I saw a brief glimpse of what this could look like at my demo, albeit a product interview is less conducive to these kinds of features than a typical planning meeting. That said, the glasses also support more general-use AI features like captioning, real-time translations for over 45 languages, the ability to take phone calls, and teleprompters/notecards for public speakers. Users can also make use of the Halliday AI assistant for voice control.

Overall, this is a much more tightly focused vision for a pair of smart glasses than Halliday’s original glasses. That said, we’ll have to see if the AI features are all that useful. So far, I’ve had little success with AI wearables successfully intuiting what I actually need surfaced in a conversation, or deriving the correct conclusions for to-do items. And at $599, the Gen 2 comes with a pretty steep price tag for the average person. That’s not including any extra costs for prescription lenses. (Halliday claims it’ll be able to handle my prescription, which is -10 in one eye and -8.75 in the other with monster astigmatism. We shall see!)

The Halliday Gen 2 are available for preorder starting today with a $10 deposit and a subsequent $100 discount on the final price. The glasses are expected to ship in September.

I’ve long argued that Hollywood has simultaneously set and ruined our expectations for smart glasses. But after binge-watching two seasons of Netflix’s A Man on the Inside, this is perhaps the first time I’ve seen Hollywood, perhaps inadvertently, illustrate the biggest cultural problem with smart glasses as they stand today. 

In a nutshell, Ted Danson plays Charles Nieuwendyk, an elderly widower who finds a new purpose working for a private investigator. Armed with a pair of Ray-Ban Meta-like glasses, a voice recorder, and a smartphone, Nieuwendyk infiltrates a retirement home, and several privacy-infringing hijinks ensue as he hunts for the jewel thief robbing the retirees. Arguably, Nieuwendyk’s actions should be despicable. He’s using smart glasses — and other gadgets — to spy on an entire retirement home of non-consenting seniors and staff. Scarier still, no one ever seems to notice the glasses’ LED privacy light — which you can see in action during the first episode. 

I suspect most viewers will view the smart glasses in this show as a mere plot device. It’s an outlandish premise, and the audience is made to understand that Nieuwendyk is a good person. He’s a bored grandpa trying to solve a crime — not, as many anti-smart glasses advocates fear, using glasses to harass women and ransom compromising footage. It helps that he’s not particularly good at staying covert, but after four decades of gracing our TV screens, it’s also hard not to fall for Danson’s palpable avuncular charm. But, as a wearables expert, what I noticed was the internal conflict Nieuwendyk faced as the show progressed — the quiet moments when he decides to leave the glasses off or delete footage. And the less-quiet moments when his new friends feel betrayed by his subterfuge, unsure whether they can trust someone who so casually invaded their privacy. 

This felt more like a reflection of my own experience as a wearables reviewer. These days, I spend months decked out in “spy glasses” and always-listening gadgets, living in what I’ve dubbed “the wearable surveillance state.” 

Two weeks ago, Meta launched cheaper smart glasses without Ray-Ban branding. Meta has been releasing new smart glasses at a steady clip ever since its Ray-Ban Meta glasses exceeded expectations in 2023. Last fall, the company even released its first pair of display glasses, complete with a nigh-invisible screen on the right lens and a futuristic wearable wristband for gesture controls. However, this launch was a bit different. This time, Meta partnered with Kylie Jenner, and ever since, the hot takes have been flowing on Threads, the company’s X-like social media platform. 

“We all agree that the Meta glasses are for perverts, yes?” Namina Forna, a New York Times bestselling author, writes in a Threads post with over 30,000 likes. 

Posts like those have been rampant over the past week or so. To be fair, it’s worth noting that these glasses aren’t capable of 24/7 audio or video surveillance as some posters allege. Battery life is nowhere near good enough; using Live AI for continuous video, taking long phone calls, or recording about 10 3K videos will drain the battery in less than an hour (and that’s if you start fully charged). Even so, the fact that you can mod the glasses or take short, stealthy clips is unsettling enough to inspire visceral reactions online. Some posts advocate that women buy the glasses to clap back and surveil the creeps harassing them. Others suggest violence, snatching people’s glasses, or publicly shaming Meta glasses users by yelling “Pedophile!” or expletives. A handful of posts point to Kylie Jenner as a hypocrite for publicly stating she values privacy while becoming a brand ambassador for Meta. 

On the flip side, several Meta glasses users have posted that they’re bewildered by the sudden hate. It’s not sudden, really — the smart glasses debate has reignited in recent months thanks to recent New York Times and Wired investigations, which found that Meta has been mulling over facial recognition features for the glasses. But smart glasses fans point to accessibility use cases, or the fact that they bought the glasses to film private, hands-free videos of their children, pets, or vacations. Others note that they hadn’t thought about the privacy implications at all and are now conflicted. The more indignant posters in this camp have argued that smartphones (and other gadgets) are also surveillance devices and that no one can reasonably expect privacy in public spaces.

Social media is never a good place for nuanced debate, but that’s particularly true of Threads, which heavily rewards rage bait. That said, these posts are a succinct if crude snapshot of the cultural tension surrounding AI wearables. These past few weeks, I’ve been testing the Ray-Ban Meta Optics, the new Meta Glasses, and Vocci, an AI note-taking ring. These posts ring in my head as I wear and use these devices in public. Because while smart glasses are at the forefront of the privacy debate, AI wearables are a category that already encompasses pendants, pins, and rings that can stealthily record conversations. Most are intended for business meetings or lectures — automatic transcripts and AI summaries are appealing to lecturers, lawyers, doctors, students, and journalists. The catch is that once a recording gadget is in my hands, I don’t have to use it as intended.

As a journalist, the Vocci ring was immensely valuable this past tech conference season. I was able to get on-the-record interviews and demos without having to use my phone. It made interview subjects a bit more natural, even though I’d explained beforehand that I was using a ring to record them and showed them the not-bright-enough recording light. While covering the Enhanced Games, I used it to discreetly record voice notes. It was immensely helpful to press the button, which would “highlight” anything said in an interview that I found interesting in real time. It was comforting to me that this wasn’t an always-on recording device like some other AI wearables I’ve tested. As an intentional work tool, where I am obtaining consent, the ring was excellent. 

Less excellent was how easily I could surreptitiously record my spouse, trusted friends, and colleagues without them noticing. (This was for testing purposes. I always disclosed and immediately deleted any such recordings afterward, while also asking for opinions on the LED recording light.) Honestly, it was so effective, I’d be shocked if AI rings didn’t feature in future spy movies or seasons of A Man on the Inside. 

Unlike Charles Nieuwendyk or James Bond, I am not a fictional character played by a charismatic actor. When I’m going about my business wearing these wearables, the public doesn’t know that I’m a tech journalist who has to adhere to a strict ethics policy. To passersby, it doesn’t matter what I do to obtain consent or the hours I spend considering how to test these devices as ethically as possible. It doesn’t matter that I’ve rehearsed and prepared for what to do and say if someone ever confronted me about the devices I’m wearing every time I walk out the door. They don’t know me, and so why should they trust me? After testing these wearables and covering this space, I wouldn’t blame anyone for mistaking me as a spy. 

The central conflict of AI wearables is that to be effective, they have to be discreet and, to a degree, covert. The Meta glasses and Vocci ring work because they’re easy to overlook. You feel comfortable wearing them, and you know you’re not being a privacy-invading glasshole. But this hinges on everyone assuming your good intent. The other sad truth is there’s nothing any individual user can do to increase trust in these devices. 

The only “successful” example I can think of is the AirTag. The same privacy and security arguments apply to AirTags as smart glasses. Surveillance gadgets have existed for ages, but the affordability and increased wearability is new and, therefore, concerning. Sure, the overwhelming majority of people use AirTags to find lost items and smart glasses for benign content recording and media playback. The bad actors who’ve misused these devices are a tiny minority — but they exist and have done undeniable harm. 

The main difference is that AirTags have an obvious killer use case, and Apple nerfed the devices by implementing unwanted tracking alerts. It doesn’t prevent misuse, but it does make it harder and less appealing. Apple has also taken feedback from domestic abuse advocates after critics pointed out the alerts weren’t robust enough. And as scary as it is, there have been videos on social media where people did find out someone was potentially stalking them before harm was committed.

There is no equivalent for AI wearables at this moment. Not to beat a dead horse, but it’s been proven that LED recording lights are easy to overlook, are washed out on a sunny day, and can be tampered with. The obvious solution to me would be to sacrifice a bit of style for harder-to-ignore signifiers. I’m talking audible shutter noises, like smartphones have in Japan — a feature required by law to prevent perverts from taking nonconsensual photos of women. Perhaps even physical shutters or miniature lens caps like DSLR cameras. One current example that comes to mind is Xreal’s modular camera attachment. You can insert the camera when you want to take photos and completely remove it when you don’t. 

“We’ve considered [a] modular [approach],” Alex Himel, Meta’s VP of wearables, told me at the Meta Glasses launch event, noting that it does come with several benefits. For example, it’d reduce the number of SKUs and designs needed. People could upgrade regular glasses easily. 

The problem, Himel says, is the glasses “wouldn’t be as clean or integrated. [They’d] be heavier, clunkier, and wouldn’t look as good.” He went on to say that Meta is aware of increasing attempts to tamper with the privacy lights and to expect more robust privacy updates soon.

After months spent as an unintentional wearable snoop, I’m convinced the AI wearable surveillance state isn’t inevitable. These are still niche devices. It’s still very early days. Enough controversy, and this could all come crashing down like a house of cards. (Just ask Google.) For all the hate on Threads — oh, the delicious irony of this discourse happening in Meta’s own backyard — there’s still time for Big Tech to put on their big boy pants, make like Spike Lee, and do the right thing. After all, it’s in their best interest. If they won’t be proactive, legislators will eventually catch up. Private companies and venues are already starting to restrict these glasses and deny entry to Meta glasses wearers. Zenni Optical is already selling anti-facial recognition lenses. 

But even if I do occasionally find this tech useful — sometimes even delightful — I’ve come to the same conclusion as Charles Nieuwendyk. Even if you have the best intentions, it feels pretty lousy being entrusted with strangers’ privacy. In TV comedies, forgiveness comes easy because conflicts are always neatly wrapped up by the season finale. That’s not guaranteed in the real world. Like Nieuwendyk, I’m just doing my job. The difference is, I never wanted to be a spy.

Correction, 7/6: A previous version of this article said Danson wears Ray-Ban Meta glasses in the show. While they have similar features and functions, they are not Meta glasses.

This is Optimizer, a weekly newsletter sent from Verge senior reviewer Victoria Song that dissects and discusses the latest gizmos and potions that swear they’re going to change your life. Opt in for Optimizer here.

About three years ago, a doctor told me I had to lose abdominal fat. She didn’t care about my lower belly fat. That, she said while pinching me, was subcutaneous fat. What she wanted was for me to target about five pounds of fat loss above my belly button because that’s where the visceral fat lived. Not weight loss; my BMI and weight, she said, were fine. I needed fat loss because I had borderline high cholesterol and one mildly elevated liver enzyme. Given my diagnosis of polycystic ovary syndrome (now known as polyendocrine metabolic ovarian syndrome), she was convinced that Type 2 diabetes lurked just around the corner for me unless I immediately overhauled my body composition. 

She suggested I invest in a smart scale — as well as commit to a vegan diet. I ignored the latter (and, a few months later, switched doctors after she recommended a dubious dieting podcast). But I did start dabbling a bit more seriously with smart scales, embarking on a multi-year journey of frustration.

Smart scales use a method called bioelectric impedance analysis (BIA) to give you a breakdown of your body fat, lean mass, and bone mass. (Some high-tech smart scales include other metrics, but it depends on the device.) The gist is that a weak electrical current is sent through your body. Fat, muscle, and other types of tissue all have different levels of conductivity, so depending on the resistance measured, an algorithm then estimates how much of each you have. They’re more accessible for consumers than other clinical methods, but they can be notoriously inaccurate. It’s very easy to have two BIA scales and get wildly different results.

Case in point, this morning I stepped on three separate smart scales: two from Withings (one only measures through your feet; the other has more advanced segmented body analysis), and a third from a company called Twin Health. When I go for my monthly doctor follow-ups, I get measured by an InBody scale, which is a fancier version of the smart scales people have at home. Two weeks ago, I went to a sports clinic to get my first Dual-Energy X-ray Absorptiometry (DEXA) scan, which shoots low-power X-rays through your body to provide a breakdown of your body composition, including bone density. It’s often considered the “gold standard” in clinical settings. None of these devices has ever given me the same body fat percentage. 

In fact, here are my most recent readings for each scale:

• Withings Body Smart: 27.4 percent
• Twin Health Scale: 28.8 percent
• Withings BodyFit: 34.1 percent
• DEXA scan: 39.3 percent
• InBody scan: 44.4 percent
  

Granted, I couldn’t take all of these measurements on the same day, so some variance is expected. However, they were all taken within two weeks of each other, after fasting, and first thing in the morning. All of these devices claim a degree of accuracy. Withings claims its BodyFit scale’s tech — which involves a retractable handle and a total of eight electrodes — underwent an internal clinical study of 80 participants and found a 99 percent correlation with DEXA scans. Let’s say the DEXA scan is the most accurate. I’m looking at a variance of 5 to 12 percentage points between it and the other devices. My highest and lowest readings differ by 17 percentage points. That is a significant difference.

And that’s just one metric. My metrics on lean mass (everything but fat) also wildly differ. DEXA, the Withings BodyFit, and InBody break down your body composition based on body parts, like torso, legs, and arms. Those readings also differ by 5 to 8 percentage points. Some also estimate bone density and visceral fat. Would you believe my measurements have some significant variance there, too? 

For example, the DEXA scan says I have extremely excellent bone density and that my skeleton makes up about 4.1 percent of my total body weight. Withings says 3.7 percent. On a scale of 1 to 20, the two Withings scales estimate my visceral fat at 1.9 and 2.1, respectively. (1 to 12 is fine, 13 to 20 is high risk.) InBody estimates at 14. My DEXA scan says I have 1.24 pounds of visceral fat and that, ideally, I’d get that number to one pound or less. 

Am I good? Frankly, that depends on who you ask. I’m sure the most hardcore fitness influencers would say that I’m obese and in dire need of locking in and shaping up. DEXA and InBody have both given me a C- score for my body composition. Ten years ago, I probably would’ve had immense health anxiety and spiraled trying to find which of these devices was “most accurate” and what numbers were “most true.”

The real truth is, it doesn’t matter all that much. 

I don’t have long-term DEXA results. Those scans are expensive (ranging from $75 to $300 a pop), and generally, it’s recommended you only take those every three to six months. But the scan was something I did to prove a point: the accuracy of these devices doesn’t matter as much as the consistency. DEXA scans are highly accurate, with a clinical margin of error of about 1 to 2 percent — but it also depends on the facility you go to, machine calibration, and your hydration levels. The first DEXA scan takes stock of where you’re at. The subsequent scans are where you start to see useful data about whether you’re progressing. That’s true regardless of which body composition method you choose — be it a pricey clinical measurement or the cheapest-possible smart scale you copped during a Prime Day sale. So long as each method is “inaccurate” by roughly the same margin of error each time, the most useful data they provide is your long-term trends. 

Again, none of these devices ever recorded the same numbers for me. Even something as simple as my weight could vary by one- to three-tenths of a pound, all other factors controlled for. But what they all did do was tell the same overall story with my trends. Body fat and visceral fat percentage uniformly went down by a significant amount. Lean mass percentage either stayed flat or increased slightly. Of the methods that measured skeletal muscle, that slightly decreased. My bone density has remained the same. Overall, these are all signs my treatment plan is working and point to the same overall recommendations: prioritize rebuilding lost muscle, continue monitoring side effects, review my blood tests every quarter, and keep on keeping on until my doctor is happy with where I’m at. 

This concept also applies to other wearables, like smartwatches and more basic fitness trackers. The daily numbers don’t mean nearly as much as your weekly and monthly trends. If you want to move more, it doesn’t matter if an Apple Watch says you took 9,000 steps and the Whoop says you took 8,000. Fretting over which one is “more right” misses the point entirely. What matters is if whatever tracker you choose measures the same mile-long walk as roughly the same number of steps (or GPS distance) each time. 

There are nuances to this. I started taking body composition analysis more seriously three years ago because a doctor scared the bejeezus out of me. Since then, I’ve experimented with about eight different scales and body composition tech of varying degrees of smartness. Until my DEXA scan, I never had a reliable sense of what my body fat percentage was. I still kind of don’t. Here’s what I do know: My body composition didn’t respond to traditional lifestyle interventions. Turns out, I have insulin resistance. That makes it easier to put on fat, which in turn increases insulin resistance. I’d never have learned that from any BIA gadget! Fixating on any individual metric — be it from a wearable, scale, or a blood test — was never going to be the key to unlocking my metabolic dysfunction. I was never going to self-optimize my way out of unlucky genetics. 

Health and fitness-wise, I’ve felt better. I’ve also felt much worse. All my numbers right now paint a very mixed picture. I could get lost in the weeds if I’m not careful. But the overall trend? I’m on a better track than I was, and that’s what’s most important.