A max-severity vulnerability in the latest Python FastAPI version of the ChromaDB project allows unauthenticated attackers to run arbitrary code on exposed servers.

The flaw is tracked as CVE-2026-45829 and was reported to ChromaDB on February 17. It received the maximum severity score from HiddenLayer, the company that discovered it.

ChromaDB is an open-source vector database and AI retrieval backend used in agentic AI and related applications. It enables retrieving semantically relevant documents during large-language model (LLM) inference.

The flaw affects the codebase containing the vulnerable Python API server logic, so the PyPI package, which has nearly 14 million monthly downloads, is at risk when servers are accessible over HTTP.

Users who deploy it locally without exposing the API server online along with those using the Rust front-end, are not affected by CVE-2026-45829.

According to HiddenLayer, a vulnerable API endpoint marked as authenticated allows attackers to embed model settings before authentication is checked.

An attacker can send a crafted request to force ChromaDB to load a malicious model from the Hugging Face platform and execute it locally. The authentication check is only performed after that step, bypassing security.

“The authentication is not missing, [it’s] just in the wrong place,” explains HiddenLayer.

“By the time it fires, the model has already been fetched and executed. The server rejects the request, returns a 500, and the attacker’s payload has already run.”

Exposure and mitigation

The researchers report that the flaw was introduced in ChromaDB 1.0.0 and was unpatched in version 1.5.8. Two weeks ago, the maintainer released version 1.5.9. However, it remains unclear if the security issue has been fixed.

Since February 17, HiddenLayer researchers have attempted to contact the developer multiple times over email and social media, but received no reply.

BleepingComputer contacted the Chroma team about the status of CVE-2026-45829 but had not received a response by the time of publication. We will update this article if additional details become available.

According to their queries on Shodan, roughly 73% of the internet-exposed instances are running a vulnerable version of Chroma.

Until it becomes clear that CVE-2026-45829 has been patched, the recommendation for impacted users is to pick the Rust frontend for their deployments or avoid exposing the Python server publicly. Another mitigation is to restrict network access to the ChromaDB API port.

The researchers also recommend scanning ML model artifacts before runtime because loading public models with ‘trust_remote_code’ effectively means executing untrusted code.

Automated pentesting tools deliver real value, but they were built to answer one question: can an attacker move through the network? They were not built to test whether your controls block threats, your detection rules fire, or your cloud configs hold.

This guide covers the 6 surfaces you actually need to validate.

Download Now

The Insta360 Mic Pro is not the smallest wireless mic you can get, but it might be the best for sound quality and certainly for build quality.

Your first impression of the Insta360 Mic Pro may be wrong. The two microphones in the set I reviewed come with what look like bright yellow labels, one of which is a smiley emoji, and you have to think you’d never wear this on camera.

But while it’s true that microphones should be unobtrusive, this one actually is. That’s because the yellow labels are not labels at all; they are e-ink displays that can be changed to anything.

So you can very easily replace them with, say, a company or show’s logo, and that new image will stay there until you change it again. I’ve just replaced both with a totally blank, black image so that the e-ink display blends in more with the rest of the microphone.

I’m okay to do this, though, because I tend to use a wireless mic on my own. When I next interview someone, it might be more useful to give their mic a different image to distinguish it.

They look like irritating stickers, but they’re actually superb color e-ink displays you can change at will.

Not that I’d admit this to anyone but you, but I am a little tempted to go find a photo of a “Star Trek” combadge to put on there.

Insta360 Mic Pro review – what you get

Insta360 Mic Pro is from the makers of Insta360 cameras, and the software reportedly makes it particularly easy to set it up for those. I tested it entirely with an iPhone, however, and actually it’s hard to imagine the setup being any easier.

Except if you do have an Insta360 X5, X4 Air, Insta360 Pro 2, or an Insta GO Ultra, the microphones will connect to them directly. For everything else, such as an iPhone, you need a receiver, which comes in the set.

The set also comes in various versions, depending on the number of microphones and receivers you choose. Insta calls microphones transmitters, and it is possible to buy one on its own, either as a replacement or as an addition.

The Insta360 Mic Pro can pair with up to four receivers, which means you could plug a receiver into four cameras and have them all recording equally good audio. Additionally, each receiver can pair with up to four microphones, allowing it to capture four interviewees on separate audio tracks.

Normally, you would get either a set with a receiver and several microphones. There is a set that has one microphone and one receiver, but I tested the edition with two microphones. In that set you get:

• 2 x Insta360 Pro microphone transmitters
• 1 x Insta360 Pro receiver
• 2 x clips for the microphones
• 2 x magnetic backplates
• 1 x charging case
• 1 x iPhone adapter
• 1 x dead cat wind shield
• 1 x carrying pouch
• 1 x USB-C to USB-C charging cable

It is all exceptionally well made. The charging case is just hard enough to open that you’ll never do it accidentally, for instance, and the receiver has a particularly fine little screen.

As shipped, the microphones come with the shirt or blouse clip attached, and the clips are so well connected that you do fear breaking something when you try to remove them. After you’ve removed it once, though, fitting and removing the clip feels straightforward.

It’s a little hard, however, slipping the microphones back into the case until you adjust the position of the clip. Alternatively, you can leave the clip off, and there is a separate slot for it within the case.

The charging case with only the receiver in it. Note the slots for the microphones, clips, and magnetic backplates.

Advertisement

Each of these elements comes individually wrapped in protective material so you can choose whether or not to load up your charging case with them.

That does include the iPhone adapter. This is a very small piece of plastic and metal that has USB-C at one end and a presumably proprietary connector at the other.

This connector slots into the base of the receiver, which initially comes with a blank plate. It all looks delicate, and you wouldn’t want to be constantly removing and perhaps losing the connector.

Except if you are solely or chiefly going to be using the Insta360 Mic Pro with an iPhone, you can fit this connector once and leave it. The receiver’s slot in the charging case has room for it.

Since this iPhone adaptor is USB-C, you can of course connect it to iPads and Macs. However, the shape of the receiver and its size are such that you can’t connect it, for instance, to the front USB-C ports on a Mac Studio without a dock.

Insta360 Mic Pro review – setting up for one and two users

When it is plugged in, the whole receiver stands a little proud of the iPhone. It’s a much thicker receiver than I’ve used before with my previous favorite mic, the tiny Hollyland Lark M2S.

Consequently, I’ve taken to making the receiver the last thing I attach after my iPhone is mounted, and the first thing I remove before I take everything down again.

In use, that receiver juts out from the iPhone enough that its exquisite little screen is clearly visible, showing you audio levels during recording. The first time you use it, you have to pair the microphones to the receiver, and an on-screen QR code prompts you through downloading the Insta360 app.

You can set a custom image on your microphone’s screen or use one of the very many supplied ones. I’m not judging.

Exactly as I started to write this, that app popped up a notification on my Mac, promoting some summertime offer or other. That’s despite the app being only on my iPhone, not on the Mac.

“Your perks have landed,” it said, and I said something very different in return.

Overall, the app is clean and simple; it was a straightforward process to download it from the App Store and pair each of the two microphones. But you do have to bat away an ad or two on the way.

You may not use the app much at all, though. As well as the essential pairing, there is what I’d call the equally essential option to get rid of the bright yellow smiley faces on the mic.

But otherwise, there are a few settings that you will come back to. Hang on a second while I tell my iPhone not to allow notifications from it.

Insta360 Mic Pro review – sound quality

I’m not an audiophile, and it would be good if there could be a “but” about here in this sentence, yet there isn’t. I do make a lot of video and audio, though, and to my ear, the Insta360 Mic Pro sounds excellent.

Top: audio from the new Insta360 Mic Pro. Bottom: the same audio recorded on a Hollyland Lark M2S

Advertisement

This is a screengrab of unedited, unaltered audio waveforms in Logic Pro. The top track is from the Insta360 Mic Pro, and the bottom one is from the Hollyland Lark M2S, which I have been very happy with up to right now.

The two audio tracks were recorded simultaneously, both microphones being worn on the same shirt and approximately equally positioned. Later in the production process, I would typically compress that Hollyland Lark M2S track and never be remotely disappointed with it, but the Insta360 Mic Pro is definitely louder and clearer without any processing.

I would like the microphones to be smaller, and I do miss that from the Hollyland ones, but the sound quality is so much better than what I’ve been using that I’ve switched to the Insta360 Mic Pro. I expect, based on the specifications, that would have made that switch even if I’d been using the same firm’s previous model, the Insta360 Mic Air.

That’s because while the Insta360 Mic Air costs less and is also quite a bit lighter at about 8 grams instead of almost 20 grams, it has a single microphone. The Insta360 Mic Pro has a three-microphone array with more pickup options.

This new model also has greater noise cancellation, and it comes with 32GB of internal storage. So the microphones themselves are backup audio recorders.

The receiver (pictured atop the provided Quick Start Guide) is larger than some, but capable of handling four audio streams

As it happens, I’ve never used an Insta360 camera. But if they’re as well-built as these microphones, I’m now very tempted.

Insta360 Mic Pro review – should you buy

I’m hardly going to say no, you shouldn’t, not after practically raving about the Insta360 Mic Pro. There is a cost for all of these points, though, and it’s enough that it can’t be a casual purchase.

But while there is the lower-cost Insta360 Mic Air (typically on Amazon for about $55), the new model is significantly better. Plus, you can buy a single microphone and receiver at first, then later add on further mics as you need them.

Insta360 Mic Pro review – Pros

• Very good audio with a three-mic array
• On-microphone backup storage
• Excellent for multi-cam and multi-interview shoots
• E-ink display on each microphone
• Extremely good build quality

Insta360 Mic Pro review – Cons

• Receiver is too large to fit in all Mac USB-C ports
• Not as small and unobtrusive as some other mics

Insta360 Mic Pro rating – 4.5 out of 5

Insta360 Mic Pro review – where to buy

The Insta360 Mic Pro set as tested here is $330 on Amazon and comes with two microphones and one receiver. A $200 single-microphone and single-receiver version will be available, although it is not on Amazon at the time of writing.

Nor is the option to buy further microphones, but that is rolling out too. A single Insta360 Mic Pro microphone costs $100 on its own.

JBL is giving its flagship Tour Series a meaningful mid-cycle refresh, but not every update lands in the same place at the same time. The JBL Tour ONE M3 wireless headphones are getting the bigger audio story with a refined sound curve based on the Harman Curve, evaluated by JBL’s Golden Ears team, controlled listening panels, and independently validated by Force Technology’s SenseLab.

The JBL Tour PRO 3 wireless earbuds are also part of the update, but with a stronger focus on user interaction and the addition of a new color option for now. In other words, JBL is not replacing its top-end personal audio lineup. It is tightening the screws, polishing the software, and trying to make two already competitive products feel more current without asking buyers to wait for an entirely new generation.

JBL Sound Curve Gets a Smarter Tune-Up

JBL is starting with the Tour ONE M3’s sound curve, which is where the most meaningful headphone update appears to land. The revised tuning still takes its lead from the Harman Curve, but JBL is presenting this as more than a quick EQ adjustment pushed through the app. The company says the update was evaluated by its internal “Golden Ears” listeners, tested through controlled listening panels, and independently validated by Force Technology’s SenseLab.

That matters because the Tour ONE M3 is already positioned as JBL’s top wireless headphone, so any change to its tonal balance has to be more than marketing seasoning. The goal appears to be a more refined and listener friendly version of the existing tuning rather than a full personality transplant. Think less “new headphone,” more “JBL tightened the alignment before sending it back onto the road.”

The new JBL sound curve debuts on the Tour ONE M3 Green Edition and will also be available to existing Tour ONE M3 owners through an over-the-air firmware update.

Firmware version 4.8.0 is designed to deliver more controlled, natural bass, along with subtle refinements to the midrange and treble. The goal is a clearer and more balanced presentation, with cleaner vocals, more believable instruments, and finer detail across the frequency range.

Pro Tip: JBL has not indicated that this sound curve update applies to the Tour PRO 3 earbuds at this time.

“The way people listen has fundamentally shifted with better hardware, higher-quality content, and smarter streaming, making today’s ears more discerning than ever. We are always striving to push the boundaries of sound performance and saw this as the perfect moment to refine the JBL sound curve for our premium headphones. The new curve makes its debut on our flagship JBL Tour ONE M3, designed to deliver the clearest, most natural listening experience we’ve ever created,” said Carsten Olesen, President of Consumer Audio at HARMAN.

Smarter Case and Transmitter Controls

JBL introduced its Smart Charging Case with the Tour PRO 3 in 2024, giving users direct access to key earbud controls without relying entirely on the app. For 2026, JBL is updating that interface on the Tour PRO 3 case, along with the JBL SMART Tx transmitter used with the Tour ONE M3 headphones.

The revised interface uses a combined horizontal and vertical menu layout designed to make favorite menu tiles easier to reach. JBL has also refreshed the graphics, enlarged the icons, updated the typography, and revised the color scheme to make navigation clearer and faster.

The goal is smoother access to advanced features, including source switching and the ability to join or create Auracast broadcasts. 

New Deep Green Finish

JBL is also adding a new color option for both the Tour ONE M3 headphones and Tour PRO 3 earbuds. The new deep green finish with copper accents takes inspiration from heritage British motorsport and current luxury design trends, giving both models a more distinctive look than the usual black, silver, or beige suspects.

JBL is positioning the finish as a style-forward option for listeners who care about performance and precision, but also want their wireless headphones or earbuds to look a little less anonymous. Not everyone wants their gear to disappear into the background.

Comparison

The Bottom Line 

JBL’s 2026 Tour Series updates are more about refinement than replacement, which is probably the smarter move in a wireless headphone and earbud market that already feels like rush hour with better noise cancellation. The biggest change is the new JBL Sound Curve for the Tour ONE M3, based on the Harman Curve and delivered through firmware version 4.8.0. It debuts with the new Green Edition and will roll out over the air to existing Tour ONE M3 headphones, bringing more controlled bass and subtle midrange and treble refinements.

The Tour PRO 3 earbuds do not appear to receive that sound curve update at this time. Their update is focused on user interaction, with a refreshed Smart Charging Case interface that mirrors the improved experience coming to the JBL SMART Tx transmitter for the Tour ONE M3. Both products also get the new deep green finish with copper accents. JBL is not launching replacements here. It is sharpening two current flagship models with better controls, improved usability, and, in the case of the Tour ONE M3, a more refined sound profile. 

Price & Availability

Related Reading:

• Google introduced Gemini Omni Flash
• It aims to make video creation easier by letting users refine projects naturally, rather than using editing software
• It’s emphasizing transparency and safety through AI watermarking and identity protections

Google’s next big AI move is aimed squarely at creativity. The company has introduced Gemini Omni at Google I/O 2026 as part of its massive slate of new Gemini features.

Omni is supposed to combine Gemini’s reasoning abilities with media creation tools that can generate and edit content across different formats.

from the some-transparency-for-take-it-down dept

Today marks the deadline for online platforms to implement a process for notice-and-takedown of nonconsensual intimate imagery (NCII) under the TAKE IT DOWN Act (TIDA), which became law one year ago. Starting today, platforms must conspicuously offer a notice-and-removal process for NCII, remove reported material within 48 hours of a “valid removal request” from the person depicted (or their authorized agent), and “make reasonable efforts to identify and remove” duplicates. (I have some qualms about the constitutionality of that last requirement, but that’s a post for another time.) 

Many members of civil society warned Congress while the bill was being negotiated that these takedown requirements are ripe for abuse, but they were ignored. Now that the provisions are in effect, we deserve to find out whether those warnings come true. Platforms should add TAKE IT DOWN takedown statistics to their periodic transparency reports.

A short refresher on the law: TIDA criminalizes the knowing and intentional disclosure of NCII, whether it’s real or AI, whether it’s of adults or minors. The criminal provisions apply to users; the takedown provisions apply to platforms. The definition of a “covered platform” encompasses public-facing user-generated content (UGC)-driven platforms, as well as sites devoted to NCII (what used to be called “revenge porn” sites). The definition exempts ISPs, email service providers, and services that mostly serve “preselected” content and to which UGC is incidental – for example, this site, which posts articles like this one but allows the Techdirt community to comment on them. 

TIDA’s takedown provisions are, in some ways, the codification of existing practices platforms already employ for removing abusive content. Some platforms have ostensibly been removing reported NCII pursuant to a voluntary initiative that dates back to 2021. There’s a similar initiative for terrorist content. Also, existing federal law criminalizes child sex abuse material (CSAM) and requires prompt reporting of apparent CSAM once a platform becomes aware of it. That law explicitly doesn’t require platforms to affirmatively go looking for CSAM (for the same constitutional reasons that give me pause about TIDA’s duplicate-removal provision). Nevertheless, numerous platforms voluntarily have tools and processes in place to proactively detect CSAM, then remove and report it. Plus, just about every online platform has notice-and-takedown processes for copyright-infringing UGC. That’s because everyone wants to qualify for the Digital Millennium Copyright Act (DMCA) safe harbor against (potentially ruinous) infringement liability.

Given these existing endeavors in the domains of NCII, CSAM, terrorism, and copyright, hopefully a lot of platforms were able to spend the past year adapting and extending their pre-existing notice-and-removal flows rather than having to reinvent the wheel. Whatever efforts they’ve put into compliance to date are about to be put to the test. Platforms should tell us – their users, Congress, the American public, NCII victims, etc. – how they’re working out.

Much has already been said here at Techdirt about the problems with TIDA’s takedown requirements, so I need not repeat those critiques at length. In brief: 48 hours is incredibly fast, the law doesn’t require a process for the user whose content was removed to appeal a takedown or have it restored, it imposes no penalties for bad-faith takedown requests, and it does a poor job of respecting First Amendment protections for speech. It immunizes platforms from liability for removing content that isn’t actually illegal NCII, while simultaneously giving the Federal Trade Commission (FTC) the power to police compliance, as the agency’s chairman reminded a dozen or so major companies in a letter last week. 

This is a recipe for rampant abuse. It incentivizes a “remove first, ask questions never” approach. And it’s particularly dangerous in our current political moment. Not only did President Trump vow to use TIDA against unflattering online speech about him, the Trump FTC is led by two hard-right Republican commissioners who’ve been using their position to pursue an anti-LGBTQ, anti-porn agenda. TIDA’s extremely abusable takedown mandate thus poses a huge risk to online free speech in general, and to content posted by queer and trans people and sex workers in particular. 

With all that said, the takedown requirements have the potential to do a lot of good. I’ve spent the past several years studying AI-generated CSAM, particularly the use of “nudify” apps – and, more recently, Grok –- to create nonconsensual deepfake pornography. I know from my research that one of the harms experienced by victims of NCII (whether deepfake or real) is the fear that anytime someone looks them up online, their NCII will come up in the results. Not only is that humiliating, it could impact their educational and career opportunities, romantic prospects, and other relationships. If platforms have to take down their NCII at their request and keep it down, that may help assuage those fears. (That said, generative AI’s capacity to rapidly make a variety of distinct images could make it hard for victims to keep up, since the onus to submit removal requests is on them or their authorized agents. That’s undeniably burdensome, even if it’s the only way for platforms to know for sure that a particular image is NCII and not constitutionally-protected consensual adult pornography.)

In short, the notice-and-takedown process mandated by TIDA might turn out to be really helpful to NCII victims, or it might be wildly abused, or both. The platforms receiving and processing those notices are the only ones who will know. That’s why they should tell us in their transparency reports. 

To be clear, TIDA doesn’t mandate transparency reporting (though Congress occasionally proposes it). Rather, transparency reports have become a standard practice by platforms over the past 15+ years; they’re typically issued once or twice per year (or even quarterly). You can review the transparency reports from companies like Google (which pioneered the practice in 2010), TikTok, LinkedIn, OpenAI, etc. Transparency reports commonly list statistics for content removals, account actions, and/or regulatory reporting for such categories as copyright, CSAM, spam, scams, and government requests to remove data or to produce data about users. There may be further breakdowns such as percent of requests fully or partially complied with, and the report may explain the platform’s policies for evaluating requests. 

Now that TAKE IT DOWN is fully in effect, platforms should add a category to their transparency reports for TIDA takedown requests, encompassing statistics as well as policy explainers. Here are a few; I’m sure veteran platform employees could add more:

• How many TIDA notices did the platform receive during the time period in question? 
• What percentage of notices did the platform comply with, and what percentage did it reject as invalid? 
• What percentage of notices involved adults, what percentage involved minors, and what percentage are age-unknown?
• What was the average time to takedown? (Of course, the platform probably won’t publish this stat unless it’s under 48 hours.)
• How many takedowns were later reversed and put back, if any?
• What is the total number of unique individuals for whom takedown notices were submitted?
• For companies acting as authorized agents, what are their names and how many requests did they submit? (Sure to be a growth industry under TIDA.)
• How does the platform count notices that fall under multiple legal authorities? For example, a minor’s nude selfie could potentially fall under 3-4 different transparency reporting categories: TIDA, CSAM, DMCA, or terms of service (TOS) violation.
• What is the platform’s policy for evaluating the validity of notices? 
• Does the platform have a process for appeals and putbacks of removed content, even though TIDA does not mandate one?
• What happens to accounts that receive multiple TIDA notices? Is there a “strike system” like many platforms have for repeated copyright infringement or TOS violations?
• What is the platform’s policy about nonconsensually-posted sexually suggestive content? Will it remove it under TIDA, or as a TOS violation? (Things like bikini or lingerie pictures technically don’t meet the definition TIDA uses for intimate imagery, but TIDA requests will surely get used for, say, the nonconsensual deepfake bikini pics Grok was churning out at the start of the year.)

This information will give insights to Congress, users, and the public about how well or poorly the TAKE IT DOWN Act is achieving its intended goal of helping victims get their nonconsensually-shared images offline, while also revealing the prevalence of the sorts of improper and abusive takedown notices the law’s shoddy drafting invites. It might also reveal tensions in how TIDA interacts with other statutes, such as Section 230, the DMCA, and CSAM laws. Those insights, in turn, could be used to make reforms to TIDA – whether to fix the deficiencies everyone already warned Congress about, address any revealed incompatibilities with other laws, or improve the law’s viability as a remedy for NCII victims (without further eroding free speech protections). 

The first half of 2026 ends six weeks from now. That might be too soon to add TIDA statistics to platforms’ H1 2026 transparency reports – especially since many platforms will have just gotten their TIDA processes up and running and will now be busy working out the kinks. But by the end of 2026, I don’t think this is too much to ask. Congress has proposed a seemingly endless number of online safety bills in recent years (RIP Mike’s blood pressure), but the TAKE IT DOWN Act is one of the few to have actually become law, warts and all. We deserve to know how it’s working.

Filed Under: csam, ftc, ncii, removals, take it down act, transparency

Last year at Google I/O, we got a promising, if frustratingly limited, look at Android XR. At this year’s event, the company confirmed that the first glasses from Warby Parker and Gentle Monster are finally coming later this year.

Those frames are still under wraps, though we got a bit of a preview during the I/O keynote. But Google’s developer conference did, at least, give us a much clearer picture of how its smart glasses will work. Given that Meta has a years-long headstart, Google will have a lot to prove. But despite being almost embarrassingly late to the smart glasses game, Google has a few significant advantages. And, after trying out the latest Android XR glasses, I suspect at least some people will prefer these over Meta’s Ray-Ban shades.

The glasses I demoed were not the branded frames briefly shown off during the keynote. They were “reference hardware” that Google uses for its own internal development. These glasses also had a built-in display, unlike the Warby Parker and Gentle Monster specs, which will be audio-only. But they didn’t really look or feel like a prototype either. While not quite as polished as my Ray-Ban Meta frames, they weren’t overly thick or nerdy looking. And they felt much lighter than the extra chunky Meta Ray-Ban Display frames.

Igor Bonifacic for Engadget

The display setup is similar to the prototype I saw last year, with a single window over the right lens. On the reference hardware, it had a 20-degree field of view, though Google was quick to point out that specific specs could change.

While the display was impressive — it was every bit as crisp and bright as the Meta equivalent — it was obvious that even the audio-only Android XR glasses could have a big advantage over Meta and other would-be rivals. Namely, that Google has been able to integrate its own apps and, yes, Gemini into the frames in a way that seems incredibly useful.

For example, using Google Translate was much more seamless than my experience with Meta’s glasses. I was able to move between a Google rep speaking Spanish and my colleague Igor Bonifacic speaking Serbian and the glasses kept the translation going without interruption. It also adeptly ignored the people around me speaking English, and only showed a real-time translation of what was being said in a foreign language. There probably aren’t that many real-life situations when you would need to move between multiple languages inside of the same conversation, but the fact that it was possible underscores Google’s advantage.

While I’ve generally been impressed with Meta’s AI glasses’ translation abilities, you can only do one language at a time. You also need to download the language to your phone ahead of time, which can make spur of the moment translations tricky.

Igor Bonifacic for Engadget

The non-display glasses will also still benefit from multimodal capabilities, which rely on the onboard cameras and Gemini to surface information based on your surroundings. I was able to look at a recipe and ask Gemini to add the ingredients to my shopping list on Google Keep. Gemini actually briefly struggled with the command, but I didn’t have to stop and start over. I kept speaking and it was able to adjust on the fly.

I’ve often complained that one of the biggest drawbacks of Meta’s glasses is that they work with relatively few third-party apps. While Meta’s working on fixing that, for now they’re great if you want to read WhatsApp messages or Instagram DMs, but there aren’t as many options outside of the company’s ecosystem. Android XR may also be heavily reliant on Google’s own ecosystem, at least for now, but being able to access Maps, Gmail and Keep feels much more practical to my everyday life.

Maps in particular could be especially useful. In my latest demo, I was once again able to get walking directions in the display, alongside a little map view when looking down towards the ground. The audio-only XR glasses won’t have the benefit of a visual guide, but Google will still be able to provide walking directions via audio cues. You can also look at restaurants and businesses around you and ask Gemini for reviews and information. I’ve long thought that travel is one of the best use cases for smartglasses. The addition of Google Maps data is a real advantage, especially when you think about combining that with other features like real-time translations and navigation.

I still have some unanswered questions about how all this will work when it’s in a pair of glasses people can actually buy. And Google still hasn’t revealed specs or pricing for the consumer version of these glasses. But there’s already a lot to look forward to.

Public Sector

Minister demands AI becomes ‘basic expectation for all public entities’

The wave of layoffs attributable to the adoption of AI has washed up on the shores of New Zealand, which has announced an overhaul of its public service that will see the technology become a “basic expectation” for government agencies and help to make it possible to sack 9,000 staff – about 14 percent of current headcount.

Finance Minister Nicola Willis announced the job cuts yesterday, in a speech that saw her bemoan the fact that New Zealand’s government comprises 39 departments and ministries, and compared that to the 16 in Australia and 24 in the UK.

She characterized the nation’s public service as “scared of AI, slow to move to the cloud” and said it operates a “complex and fragmented set of overlapping IT solutions.”

“Our government is as frustrated as you are by the fragmentation and silos, the complexity, the status-quo thinking and the dangerously slow take up of digital and AI technologies,” she added.

Aotearoa’s answer is to task its Chief Digital Officer “to embed AI deployment as a basic expectation for all public entities.”

Minister Willis mentioned a “recent trial of an AI scribe tool in hospital emergency rooms which has reduced the amount of time clinicians have to spend on file notes and increased the time they spend with patients” as an example of the sort of thing she hopes to replicate.

She said the planned overhaul will therefore “reduce the number of government departments, increase the use of AI and other digital tools, and deliver significant savings.”

The government plans to cap departmental budgets and says that combined with redundancies it will save NZ$2.4 billion ($1.4 billion) over four years – less than one percent of all core government spending.

Plenty of tech companies have made substantial redundancies that they justify as necessary to create an appropriate workforce for the age of AI, an explanation we’ve seen deployed to explain deep cuts at Cisco, Cloudflare, Atlassian, Meta, and Arctic Wolf.

Few governments have done likewise, but one early high-profile effort – the Elon-Musk-led “Department of Government Efficiency” – hoped to use AI to improve government operations but left behind little evidence it had succeeded.

New Zealand is blessed with many resources and extraordinary natural beauty, but has a modest tax base – yet residents expect a high level of government services. Minister Willis’s plan is therefore a very big bet on AI. ®