Technology

Mattel is partnering with OpenAI to build AI‑powered toys, which might lead to some amazing fun, but also sounds like the premise for a million stories of things going wrong.

To be clear, I don't think AI is going to end the world. I've used ChatGPT in a million ways, including as an aide for activities as a parent. AI has helped me brainstorm bedtime stories and design coloring books, among other things. But that's me using it, not opening it up to direct interaction with children.

The official announcement is very optimistic, of course. Mattel says it’s bringing the “magic of AI” to playtime, promising age‑appropriate, safe, and creative experiences for kids. OpenAI says it’s thrilled to help power these toys with ChatGPT, and both companies seem intent on positioning this as a step forward for playtime and childhood development.

But I can’t help thinking of how ChatGPT conversations can spiral into bizarre conspiracy theories, except suddenly it's a Barbie doll talking to an eight-year-old. Or a GI Joe veering from positive messages about "knowing is half the battle," to pitching cryptocurrency mining because some six‑year‑old heard the word “blockchain” somewhere and thought it sounded like a cool weapon for the toy.

As you might have noted from the top image, the first thought I had was about the film Small Soldiers. The 1998 corny classic about an executive at a toy company deciding to save money by installing military-grade AI chips into action figures, leading to the toys staging guerrilla warfare in the suburbs? It was a satire, and not a bad one at that. But, as over-the-top as that outcome might be, it's hard not to see the glimmer of chaotic potential in installing generative AI in the toys children may spend a lot of time with.

I do get the appeal of AI in a toy, I do. Barbie could be more than just a doll you dress up, she could be a curious, clever conversationalist who can explain space missions or play pretend in a dozen different roles. Or you could have a Hot Wheels car commenting on the track you built for it. I can even picture AI in Uno as a deckpad that actually teaches younger kids strategy and sportsmanship.

But I think generative AI models like ChatGPT shouldn't be used by kids. They may be pared down for safety's sake, but at a certain point, that stops being AI and just becomes a fairly robust set of pre-planned responses without the flexibility of AI. That means avoiding the weirdness, hallucinations, and moments of unintended inappropriateness from AI that adults can brush off but kids might absorb.

Toying with AI

Mattel has been at this a long time and knows what it is doing, in general, with its products. It's certainly not to their advantage to have their toys go even slightly haywire. The company said it will build safety and privacy into every AI interaction. They promise to focus on appropriate experiences. But “appropriate” is a very slippery word in AI, especially when it comes to language models trained on the internet.

ChatGPT isn’t a closed-loop system that was built for toys, though. It wasn’t designed specifically for young kids. And even when you train it with guidelines, filters, and special voice modules, it’s still built on a model that learns and imitates. There’s also the deeper question: what kind of relationship do we want kids to have with these toys?

There’s a big difference between playing with a doll and imagining conversations with it, and forming a bond with a toy that independently responds. I don’t expect a doll to go the full Chucky or M3gan, but when we blur the line between playmate and program, the outcomes can get hard to predict.

I use ChatGPT with my son in the same way I use scissors or glue – a tool for his entertainment that I control. I’m the gatekeeper, and AI built into a toy is hard to monitor that way. The doll talks. The car replies. The toy engages, and kids may not notice anything amiss because they don't have the experience.

If Barbie’s AI has a glitch, if GI Joe suddenly slips into dark military metaphors, if a Hot Wheels car randomly says something bizarre, a parent might not even know until it’s been said and absorbed. If we’re not comfortable letting these toys run unsupervised, they’re not ready.

It’s not about banning AI from childhood. It’s about knowing the difference between what’s helpful and what’s too risky. I want AI in the toy world to be very narrowly constrained, like how a TV show aimed at toddlers is carefully designed to be appropriate. Those shows won't (hardly ever) go off script, but AI's power is in writing its own script.

I might sound too harsh about this, and goodness knows there have been other tech toy scares. Furbies were creepy. Talking Elmo had glitches. Talking Barbies once had sexist lines about math being hard. All issues that could be resolved, except maybe the Furbies. I do think AI in toys has potential, but I'll be skeptical until I see how well Mattel and OpenAI navigate that narrow path between not really using AI and giving the AI too much freedom to be a bad virtual friend to give your child.

You might also like

• Anthropic’s new AI-written blog is more of a technical treat than a literary triumph
• AI can write a hit song, but it can’t lift your soul or break your heart
• Audible’s AI narration sounds impressive, but I'd rather hear the story told by a human

• OpenAI has upgraded ChatGPT's Projects feature to remember past chats, tone preferences, and files
• Projects now offers deep research, voice mode, mobile file uploads, and more
• OpenAI wants Projects to function more like smart workspaces than one-off chats

ChatGPT's Projects feature has been a useful way to organize conversations with the AI chatbot since it debuted, but it has had its limitations. A major set of upgrades released by OpenAI this week has transformed Projects from a simple file folder into a highly focused version of ChatGPT as a whole.

The Projects feature debuted as a way to organize related chats and files into one digital shoebox. But now, that organization means ChatGPT will remember that those chats and files are related. So, if you start a chat within a Project, the AI will remember things from other chats in that project, referencing your past messages within the same workspace.

If you start a new Project, you can upload your notes, chat about the topic with ChatGPT, ask for online comparisons, and then come back three days later to continue the conversation without rehashing everything or having the AI cite unrelated discussions. ChatGPT won’t just remember the topics either. It will remember your formatting preferences, as well as your tone of voice.

And those can be a lot more complex conversations now that Projects includes the Deep Research tool, which lets you run multi-step tasks in ChatGPT, blending your files and instructions with live information from the web.

You can also now use ChatGPT's voice mode in Projects. Tap the microphone inside any project and start talking about the files within or anything else, and you'll see it appear. And if you're using the mobile app, you can now upload files directly and switch between GPT-4o or other models on the fly.

Other upgrades are more minor but still significant. For instance, if you have a Project that you don't want to share in its entirety, but it includes a particular ChatGPT conversation you wish to send to someone, you can do that now. And if a discussion with ChatGPT suddenly inspires you to start a Project, you can now drag it directly into a project folder or convert it instantly from the sidebar.

Not everyone can use the upgraded Projects features as of yet. You have to be a ChatGPT Plus or Pro subscriber for now. However, based on many other ChatGPT features that were once exclusive to subscribers, I wouldn't be surprised if these become accessible to free users at some point in the future.

AI project power

As impressive as ChatGPT Projects could be now, I wouldn't expect to see offices throwing out their Notion or Trello programs anytime soon. They still lack some of the common elements of those tools, like calendars. But, for personal or just smaller efforts, it's a nice enhancement of the AI assistant, one that might at least help OpenAI compete with the AI infusions Google has been adding to its ecosystem.

OpenAI has been clear that they don't just want to be a chatbot provider. They want to be your go-to for life and work. These upgrades feel like the early sketches of something more ambitious. OpenAI might someday pitch ChatGPT as an alternative to toggling between ten apps. Instead, you might one day just open ChatGPT and say, “Let’s pick up where we left off on the next work presentation.”

A little experimentation on my part found the upgraded Projects seemed more efficient almost immediately, but not without some hiccups. One large collection of conversations I've organized for testing other features was a little too eager to dredge up the initial interactions rather than pull from more recent discussions about ChatGPT's capabilities. And while Projects can now reference past chats, the actual search and navigation between those chats still isn’t perfect. There's no Boolean logic to use to isolate certain phrases yet, so you might have to do some scrolling to find what you're looking for.

Still, even with the inevitable friction, I can see the value of making Projects more of a self-organizing AI data source, rather than simply a file folder for documents, as it has been. Whether compiling research, analyzing data, or plotting the perfect party, it could make using ChatGPT a lot less chaotic.

You might also like

• ChatGPT's new Projects feature can organize your AI clutter
• How to use ChatGPT’s new Projects feature to bring order to OpenAI’s chatbot
• ChatGPT crosses a new AI threshold by beating the Turing test

• Gigabyte's AI TOP 500 TRX50 is a desktop built for AI developers working on massive LLMs
• Older Threadripper CPU included but motherboard supports newer upgrades
• GeForce RTX 5090 GPU paired with software for model tuning

Gigabyte has quietly launched the AI TOP 500 TRX50, a high-end system aimed at developers working on AI models and advanced multimodal tasks.

The machine is powered by AMD’s 24-core Ryzen Threadripper PRO 7965WX processor and cooled by an AORUS 360 AIO liquid cooler. This combination allows it to outperform Gigabyte’s previously announced Arrow Lake-S-based AI TOP 100 Z890.

Interestingly, as TechPowerUp reports, the AI TOP 500 still relies on AMD’s current-generation "Zen 4/Storm Peak" architecture, even with the Threadripper PRO 9000 series expected to launch in the near future.

Ports galore

Fortunately, Gigabyte’s TRX50 motherboard supports future upgrades, which could appeal to those planning longer-term builds. VideoCardz suggests that a version using the 32-core 7975WX might be released soon.

Like the AI TOP 100, the 500-series prebuild also includes Gigabyte’s own GeForce RTX 5090 Windforce graphics card.

Internally, the desktop (which bears a striking resemblance to the Cooler Master HAF 700) supports up to 768GB of DDR5 R-DIMM memory across eight slots.

Storage comes in the form of a 1TB AI TOP 100E cache SSD, built to endure heavy write cycles, and a 2TB AORUS Gen 4 SSD for primary use. Power is provided by a 1600W AI TOP Ultra Durable PSU rated at 80 Plus Platinum and compatible with ATX 3.1.

The AI TOP 500 offers a wide range of connectivity options. Up front, users get four USB 3.0 ports, one USB 3.2 Gen 2 Type-C port, audio in and out jacks, and both power and reset buttons. On the rear are six USB 3.2 Gen 2 ports, two USB4 40Gbps Type-C ports, dual RJ-45 LAN ports, a DisplayPort input, and two additional audio jacks.

The workstation also supports multi-node expansion through Thunderbolt 5 and Dual 10G LAN, making it a practical option for research labs or development teams.

The system is tightly integrated with Gigabyte’s AI TOP Utility software platform which helps users manage AI models, build datasets, and monitor hardware performance in real time.

With support for up to 405 billion parameter models, Gigabyte is targeting users who require serious local compute performance without relying on cloud resources…. And, gamers too, apparently, if its tagline of “Premium gaming & AI empowered desktop” is to be believed.

You might also like

• These are the best mini PC choices around right now
• And these are best mobile workstations to suit your every need
• HP's Ryzen AI Max+ Pro 395 laptop with 128GB RAM goes on sale everywhere in the US

• Forescout report finds many vulnerable solar devices run outdated firmware with known exploits active in the wild
• Europe holds 76% of all exposed solar power devices, with Germany and Greece particularly at risk
• SolarView Compact exposure jumped 350% in two years, and it's already linked to cybercrime

The rapid growth of solar energy adoption worldwide has sparked renewed concerns about cybersecurity vulnerabilities within solar infrastructure.

A study by Forescout’s Vedere Labs found nearly 35,000 solar power devices, including inverters, data loggers, and gateways, are exposed to the internet, making them susceptible to exploitation.

These findings follow a previous report by Forescout which identified 46 vulnerabilities in solar power systems.

High exposure and geopolitical implications

What’s particularly alarming now is that many of these devices remain unpatched, even as cyber threats grow more sophisticated.

Ironically, vendors with the highest number of exposed devices aren’t necessarily those with the largest global installations, suggesting issues such as poor default security configurations, insufficient user guidance, or unsafe manual settings.

Forescout found Europe accounts for a staggering 76% of all exposed devices, with Germany and Greece most affected.

While an internet-exposed solar system isn’t automatically vulnerable, it becomes a soft target for cybercriminals. For example, the SolarView Compact device experienced a 350% increase in online exposure over two years and was implicated in a 2024 cyber incident involving bank account theft in Japan.

Concerns around solar infrastructure deepened when Reuters reported rogue communication modules in Chinese-manufactured inverters.

Although not tied to a specific attack, the discovery prompted several governments to reevaluate the security of their energy systems.

According to Forescout, insecure configurations are common, and many devices still run outdated firmware versions. Some are known to have vulnerabilities currently under active exploitation.

Devices like the discontinued SMA Sunny WebBox still account for a significant share of exposed systems.

This is not just a matter of faulty products, it reflects a system-wide risk. While individually limited in impact, these internet-exposed devices may serve as entry points into critical infrastructure.

To mitigate risk, organizations should retire devices that cannot be patched and avoid exposing management interfaces to the internet.

For remote access, secure solutions such as VPNs, along with adherence to CISA and NIST guidelines, are essential.

Additionally, a layered approach using top-rated antivirus tools, endpoint protection solutions, and especially Zero Trust Network Access (ZTNA) architecture may be necessary to keep critical systems insulated from intrusion.

You might also like

• Experts warn GTA and Minecraft being used to lure in cyberattack victims
• These are the best green web hosting providers
• We’ve rounded up a list of the best web hosting providers

How much more does the Pro cost? How much lighter is the Air? Which size is best? You've got MacBook questions. We've got answers.

• AMD is aggressively acquiring talent to bridge the Instinct and Blackwell GPU performance gap
• Brium’s compiler expertise could help AMD accelerate inference without hardware-specific dependencies
• Untether AI's team joins AMD, but existing customers are left without product support

AMD’s recent moves in the AI sector have centered around strategic acquisitions aimed at strengthening its position in a market largely dominated by Nvidia.

These include the acquisitions of Brium, Silo AI, Nod.ai, and the engineering team from Untether AI, each targeted at bolstering AMD’s AI software, inference optimization, and chip design capabilities.

The goal is clear: narrow the performance and ecosystem gap between AMD’s Instinct GPUs and Nvidia’s Blackwell line.

Calculated acquisitions amid a competitive ecosystem

AMD described the acquisition of Brium as a key step toward enhancing its AI software capabilities.

“Brium brings advanced software capabilities that strengthen our ability to deliver highly optimized AI solutions across the entire stack,” the company said.

Brium's strengths lie in compiler technology and end-to-end AI inference optimization, areas that could be crucial for achieving better out-of-the-box performance and making AMD’s software stack less reliant on specific hardware configurations.

While this makes for a strong technical case, it also suggests that AMD is still playing catch-up in the AI software ecosystem, rather than leading it.

Brium’s integration will affect several ongoing projects, including OpenAI Triton and SHARK/IREE, which are seen as instrumental in boosting AMD’s inference and training capabilities.

The use of precision formats such as MX FP4 and FP6 points to a strategy of squeezing higher performance from existing hardware. But the industry has already seen similar moves from Nvidia, which continues to lead in both raw processing power and software maturity.

Another notable move was AMD’s absorption of the entire engineering team from Untether AI, a Canadian startup known for its energy-efficient inference processors. AMD didn’t acquire the company, only the talent, leaving Untether’s products unsupported.

“AMD has entered into a strategic agreement to acquire a talented team of AI hardware and software engineers from Untether AI,” the company confirmed, highlighting a focus on compiler and kernel development along with SoC design.

This signals a strong push into inference-specific technologies, which are becoming increasingly critical as training-based GPU revenue faces potential decline.

“AMD’s acquisition of Untether’s engineering group is proof that the GPU vendors know model training is over, and that a decline in GPU revenue is around the corner,” said Justin Kinsey, president of SBT Industries.

While that may overstate the situation, it reflects a growing sentiment in the industry: energy efficiency and inference performance are the next frontiers, not simply building the fastest systems for training large models.

Despite AMD’s optimism and commitment to “an open, scalable AI software platform,” questions remain about its ability to match Nvidia’s tight integration between hardware and CUDA-based software.

Ultimately, while AMD is taking calculated steps to bridge the gap, Nvidia still holds a considerable lead in both hardware efficiency and software ecosystem.

These acquisitions may bring AMD closer, but for now, Nvidia’s Blackwell remains the benchmark for what is widely regarded as the best GPU for AI workloads.

You might also like

• These are the best video cards for video editing you can buy right now
• Take a look at the best virtual machine software
• Fake Cloudflare CAPTCHA page laden with malware uncovered in the wild

Here are some hints and the answers for the NYT Connections: Sports Edition puzzle, No. 264, for June 14.

Here are hints and answers for the NYT Strands puzzle No. 468 for June 14.

Unseasonable warmth coming soon to Alaska has prompted the first-ever heat advisory for residents.

• Default passwords and outdated firmware are turning your home camera into a public livestream, report warns
• Thousands of exposed webcams are offering a front-row seat into private and corporate life
• A simple web browser is all it takes to peek into 40,000 unsecured camera feeds

Thousands of internet-connected webcams, intended to enhance safety and convenience, are now unintentionally offering a window into private lives and secure environments.

Research by Bitsight claims over 40,000 webcams around the world are publicly accessible online, often without their owners’ knowledge.

These include security cameras, baby monitors, office surveillance systems, and even devices inside hospitals and factories.

A growing digital threat, not a hypothetical one

The investigation highlights just how easily accessible these cameras are.

“No passwords. No protections. Just out there,” wrote João Cruz, Principal Security Research Scientist at Bitsight TRACE, noting it requires neither elite hacking skills nor expensive software. In many cases, all it takes is a web browser and a valid IP address.

“We first raised the alarm in 2023, and based on this latest study, the situation hasn’t gotten any better.”

Exposed footage ranges from innocent scenes, like bird feeders, to far more sensitive views, such as home entry points, live feeds from living rooms, whiteboards in office spaces, and even operations inside data centers.

Worryingly, disturbing conversations have emerged on dark web forums, where some users share methods for locating exposed cameras, or even sell access to live feeds.

“This isn’t hypothetical: this is happening right now,” Cruz emphasized.

The United States leads with roughly 14,000 exposed cameras, followed by Japan, Austria, Czechia, and South Korea. These aren’t isolated incidents but part of a broader failure in how internet-connected cameras are deployed and managed.

Bitsight’s team scanned for both HTTP- and RTSP-based cameras, and the results suggest these figures may only scratch the surface.

Many of the exposed devices result from basic setup errors: default credentials, open internet access, and outdated firmware that leave systems vulnerable.

While vendors and manufacturers must improve device security, users also share responsibility.

Choosing products vetted for cybersecurity can help, but users should also pair their camera setups with tools like leading antivirus software and top-rated parental control solutions, which often include network monitoring to flag unusual access or unprotected devices.

Ultimately, private users should always check remote accessibility settings, change default passwords, update firmware regularly, and, especially for enterprises, enforce firewall protections and require VPN access.

You might also like

• These are the best VPN with antivirus solutions
• Take a look at our pick of the best password managers
• Epson launches subscription where you pay "just" $7.99 to rent a printer and print 50 pages every month

Here are hints and the answer for today's extra-difficult Wordle No. 1,456, June 14.

Here are some hints and the answers for the New York Times' Connections puzzle for Saturday, June 14, No. 734.

Cybercriminals attack retailers looking for a payday, but consumers take the hits, especially when the target is food suppliers.

Well, for the iPad faithful, Apple’s WWDC 2025 keynote was the day that faith was rewarded. I, like countless others, have been waiting for a major upgrade for iPadOS, and the Cupertino-based tech giant delivered.

Yes, iPadOS 26 brings with it Liquid Glass, but more importantly for all iPads that support it, you’ll get actual windowed multitasking, the ability to drop folders in the dock, a menu bar up-top, one of the most addicting gestures I’ve used, and the ability for tasks to run in the background.

Easily, it was the standout moment from the keynote, and I got to go hands-on briefly with iPadOS 26 running on a 13-inch iPad Pro with M4 attached to a Magic Keyboard with an Apple Pencil Pro.

Now, let’s address the elephant in the room – allowing those landmark features I listed out above makes the iPad seem like a Mac, but don’t call this a Mac. Yes, Apple did take some features from the Mac rather than reinventing the entire concept – say, for the close, minimize, or expand buttons in the top left or the menu bar – but this is all well thought out for the iPad, and takes advantage of one of the best parts of an iPad.

Multitouch.

With the iPad’s approach, it’s sort of a choose-your-own-adventure, while on the Mac, it’s keyboard and trackpad. I used it and saw a demo of fingers controlling the windows, as well as using the Apple Pencil to move items around and even the cursor. It's all about control, in that however you see it and want to use it, you can do so to get more out of your iPad. Let’s talk about why.

(Image credit: Jacob Krol/Future)

Let’s start with the most exciting part – from any app, you can pull from the bottom corner – it’s set with an effect, a slightly darker edge in the bottom right – to easily resize the window by pulling it back and forth. So from full screen, you just pull it towards the other side to make it smaller, by width or height, and then you can grab the top of the window to place it where you like.

Using the dock, you can then drag and drop another app up or do a swipe up for the peek mode to access your home screen and place any app in this layout. It’s really smooth and lets you finally have your ultimate iPad layout. Maybe that’s a Safari window open to a Google Meet in the corner, the reminders app for your checklist, and your email as you start your day.

You can also split the screen with an image and then open an app like ProCreate, allowing you to see your starting point while drawing something awesome. It really lets you tailor the experience to how you see fit.

Now this new windowing setup does replace SplitView and SlideOver, and while that didn’t excite me when I first heard, I do like the various preset options you can pick from via a long press in the top left corner of any window and the new gesture.

(Image credit: Jacob Krol/Future)

With a flick to the left or right, you can effortlessly split your screen and then adjust it further by moving the slider in the middle as needed. This feels like an easier way to achieve a similar result to SplitView, and is quite frankly fun to do.

You can also tap the top of the iPad’s screen to access a menu bar for things like more precise settings or easy exports – it’s the most similar part of the experience to the Mac. Still, considering it’s hidden until you need it, I think iPad power users will likely get the most out of this.

It feels really natural in this implementation, and not a cookie-cutter copy and paste from the Mac, given the updated elements and the ability to control with both touch and a trackpad.

(Image credit: Jacob Krol/Future)

Complementing the new multitasking approach is a significantly improved Files app and a dock that can now display a live folder. The app will feel familiar, but a new list view with the ability to customize modifiers, also known as the columns you see, will really let you tailor this for your specific needs.

For instance, I could see myself sorting by last modified and then pulling the folder containing images to the dock to edit in an app like Pixelmator, export, and then upload it into a content management system for a story build. Changes you make within folders or to these layouts can be synced across devices and updated in iCloud as well. If you’re a fan of colored folders and keen to name with emojis, you get this as well.

Those larger exports, maybe a batch photo editor or video export from Final Cut Pro, can now run in the background. I got a demo of this, and it either lives at the top of your device with a progress bar or in a little icon near your time where you can track multiple exports or tasks.

(Image credit: Jacob Krol/Future)

The really exciting part, even from these demos and a little usage, is the fact that this isn’t just limited to the iPad Pro with M4 or the iPad Air with M3 or even another step-up model. This new multitasking experience is the result of a new ‘Window Prioritization Model’ that works in conjunction with the performance and resource manager. It has been entirely re-architected to run on any iPad that supports iPadOS 26.

Meaning that the 9th Gen iPad – one of the best values Apple’s ever released – will get this new multitasking experience, same for the 10th, 11th, or 13th Gen, the iPad Air, iPad mini, and Pro. You might not be able to open a dozen there all at once, but it will let you push the chip inside further.

For now, iPadOS 26 is in a developer beta, which means it's not for your main device as bugs and issues are to be expected, but a public beta will arrive in July, and this will be released for everyone with an eligible device in the Fall. I’m super excited to spend time with it and eventually give it a full review treatment, but for now, it’s the upgrade we’ve been waiting for that feels distinctly like an iPad.

Sure, the Mac has long been the ultimate in productivity, but that lacks touch and is truly designed for keyboard and trackpad. The iPad is multitouch first, and Apple really put the time in to craft an experience that feels purpose-made for multiple inputs, with touch being first.

Just fair warning, I’ll be using many, many windows.

You might also like

• Live from WWDC 2025 – TechRadar podcast unpacks that massive iPadOS update and looks through Liquid Glass
• This is what really happened with Siri and Apple Intelligence, according to Apple
• You won't believe how insanely realistic Personas are in visionOS 26 on the Vision Pro

• The Fiio M21 hi-res audio player costs £279 / $329 (about AU$583)
• Quad-DAC configuration and two-stage amplification
• Desktop mode for optimized output for hi-fi systems

The arrival of a new affordable audio player from FiiO is guaranteed to get our attention: when we reviewed the JM21 portable music player we were blown away by its sound and its value for money. And now there's a new player that's more powerful without being much more expensive.

Where the JM21 is a dual-DAC device, the new FiiO M21 is a quad-DAC player. And double the DACs doesn't mean double the price: where the JM21 launched at $199 / £179, the M21 is $329 / £279.

FiiO M21 portable audio player: key specifications and features

(Image credit: FiiO)

The M21 is a sleek-looking thing with a big 4.7-inch IPS touchscreen on the front, a glass back, and an octa-core Snapdragon 680 inside it. The device runs Android 13 and comes with 4GB of RAM, 64GB of internal storage with 52GB of that available for use, and you can add up to 2TB via microSD. Bluetooth is 5.0 (SBC, AAC, aptX, aptX HD, LDAC and LHDC).

The DACs are four Cirrus Logic CS4319s arranged to minimize crosstalk and noise. The two-stage amplification provides voltage first and current second, which Fiio says creates a clean and dynamic sonic profile.

A key new feature is M21, a first for Fiio's entry-level Androids. It enables you to connect to a USB-C supply and bypass the battery, delivering up to 21Vpp peak voltage and 950mW per channel into 32 ohms. Desktop mode also enables you to use the M21 as a hi-res streamer for a wider hi-fi system.

Battery life in portable mode is 15 hours over the 3.5mm output and 11.5 hours with the balanced 4.4mm output. And there's an optional retro case that looks like a cassette tape – it's the larger metal-colored box the M21 is sitting on in the main photo at the top.

There's no doubt that this is going to sound fantastic – and I suspect it'll probably cope better with higher volumes than the JM21, which gets a bit shouty when you crank things up too high. And at £279 / $329 the M21 still undercuts many rival hi-res audio devices – not least Fiio's own M23, another player we love.

Between that and the fake-cassette cas,e I think FiiO's come up with a winning formula here. The player is available now.

You might also like

• Best MP3 player 2025: top portable hi-res music players for everyone
• This affordable new hi-res music player is the iPod replacement I've been waiting for
• This cheap Hi-Res Audio music player is like a modern iPod mini, with the funky colors to match

• Researchers are warning about the "unsubscribe here" button in spam emails
• They can be used to redirect victims to malicious pages
• There are other ways to get rid of spam, so users should be on their guard

If you've received a spam email with an “unsubscribe here” button at the bottom, don’t press it - it could do more harm than good.

This is according to TK Keanini, CTO of DNSFilter, who recently revealed pressing such a button sends the recipient away from the safety of the email client and into the open internet, where potentially malicious landing pages are lurking.

In fact, Keanini claims that one in every 644 clicks can lead to a malicious website.

How to unsubscribe, then?

Even if clicking the button doesn’t lead directly to a phishing page, other, more subtle, risks, are lurking as well.

Keanini says that hackers would often place that button just to see who clicks - which would also help them determine which email addresses are active and thus worth targeting further.

The general rule of thumb seems to be - if you don’t trust the company that sent the email, don’t trust the unsubscribe process, either.

So, what’s the alternative? The alternative is to unsubscribe through the email client itself, rather than through the email’s body.

Most email clients have “list-unsubscribe headers”, which appear as built-in buttons and thus don’t include source code, Tom’s Guide explained. “If your email header doesn’t contain a link, you can reply on your spam filters, or try blacklisting the sender instead,” the publication further explained.

Those who don’t have these options can use disposable email addresses when signing up for different services. Most email service providers allow users to create throwaway email addresses, as well. For example, Gmail has a feature called “plus addressing” or “Gmail aliases”, which allow users to modify their address by adding a + and a tag before the @gmail.com address.

That way, the email address used during registration could be yourname+shopping@gmail.com. Messages will still arrive in the inbox, but they can be easily tracked or filtered.

Via Wall Street Journal

You might also like

• Ransomware hackers demand victims justify their jobs, or pay up
• Take a look at our guide to the best authenticator app
• We've rounded up the best password managers

On June 17, PlayStation Plus subscribers can take on supernatural forces or an international conspiracy.

Steam Next Fest lets you get a hands-on first peek at the best upcoming new games. Here are the best demos from this Next Fest.

How is it already June when January lasted about three years? Well, somehow, we've made it to the halfway point of 2025 and the best streaming services are ramping up their offerings as we head into the summer season. With the rising cost of inflation kicking in pretty much everywhere, it's not always optimal to maintain subscriptions to Every. Single. Service.

Enter: streaming hopping. Lots of people are opting to snip 'n' save every month by pausing subscriptions based on what's available to watch.

That's why I've pored over the main streamers and what they're offering this upcoming month, to help you decide which ones to keep, and which ones to pause for June 2025. I'll be exploring why Netflix is rocking my watchlist and why its one-time biggest competitor Prime Video remains on hold.

Why I'm not canceling Netflix…. again

(Image credit: Netflix)

Look, I'm as shocked as anyone to discover that the staple streamer I always refer to as the one I keep just because, is offering up plenty of new morsels that I'm genuinely excited to watch. And that doesn't include all of the teasers that debuted at this year's Tudum event.

First up, out of everything coming to Netflix in June, let's take a peek at the library titles. If classic horror is your bag, you'll be pleased as punch as a trio of top-shelf Hitchcock titles are scheduled to arrive June 1. The Birds, Frenzy, Rear Window, and Vertigo are the perfect jumping-off point if you've always wanted to get into Hitch but didn't know where to start. I only saw Vertigo for the first time recently and can't believe how long it took me to see it! So don't be like me – watch it!

Elsewhere on the new front, we've got a bunch of shows I'm very excited for. One is arguably one of the best Netflix shows ever made: Squid Game. Every single episode of Squid Game season 3 drops on June 27, bringing to a close this hugely popular series that makes you realize that things could always be worse – we could be fighting for our lives playing schoolyard games. Since season one, this South Korean series has become one of the water cooler shows of the season, so make sure to catch up on the first two seasons.

(Image credit: Netflix)

I'm excited as heck for The Ultimatum: Queer Love season 2 that drops in its entirety on June 25. I am typically not the biggest fan of reality TV, but the first season of this took over my life for a few days back in 2023. I expect season 2 will do exactly the same.

If it's cosy you're after, then look no further: Ginny and Georgia season 3 arrives on June 5 followed by Greys Anatomy season 21 (I'm sorry, season 21??) on June 14. Sometimes you need those shows that are the streaming equivalent of a weighted blanket, and this duo fit the bill.

Speaking of cosy, a couple of recent horrors are set to arrive on June 1, so if you wanna dive into the world of Jordan Peele scope out Us. Ahead of Weapons, Zach Cregger's sophomore effort due to arrive in theaters this summer, catch up on his 2022 debut Barbarian. I recommend going in knowing nothing!

Why I'm leaving my Prime Video subscription paused… also, again

(Image credit: Prime Video )

Until Prime Video delivers a whammy of must-see movies or shows– like October 2021 when it dropped a quartet of Original horror movies – I'm keeping my subscription on pause. That's not to say it will last forever, with the streamer promising a raft of intriguing titles in the future, but for now it's feeling a bit like… everything good is fleeing!

Last month's new Kevin Bacon series recently received the axe after one season, which only dropped in May, and they also canceled the sprawling Robert Jordan adaptation The Wheels Of Time following a three season-run.

Here's hoping Prime Video pivots back soon, but until then, I'll be sticking with Netflix.

You might also like

• 5 new Netflix movies to stream in June 2025
• 5 new action movies coming to Netflix in June 2025
• Prime Video added 50 new movies in June

• Kioxia details its plans for a new SSD for AI workloads reaching 10 million IOPS
• IOPS measures how quickly a storage device can handle small, random requests
• New drive uses XL-Flash, a type of SLC NAND, and a new in-house controller

Kioxia has unveiled plans for a new SSD it says could hit an impressive 10 million IOPS, a level of performance aimed squarely at the demands of AI-driven systems.

The SSD will use XL-Flash, a type of single-level cell (SLC) NAND, combined with a new in-house controller.

A Kioxia spokesperson told TechPowerUp, “We’re taking our ultra-fast XL-Flash memory chips, which use single-level cells, and pairing them with a completely new controller… We're targeting over 10 million IOPS, and we plan to have samples ready by the second half of 2026.”

Difference between IOPS and GBps

IOPS, or input/output operations per second, measures how quickly a storage device can handle small, random requests, particularly important in AI and server applications where fast access to small files is key.

This is different from GBps, which refers to the actual data transfer speed and is used to measure how fast large files can be read or written.

A drive with high GBps might excel in video editing or large file transfers, but for machine learning tasks where thousands of small data packets are read or written constantly, high IOPS matters more.

Kioxia’s approach to next-gen storage includes not just one-off projects but a wider effort to meet varied use cases. Its CM9 series, which is sampling to customers now, focuses on speed and reliability to match high-end GPUs used in AI, while the LC9 series delivers massive 122TB capacities for large databases.

Behind these products is the 8th generation BiCS FLASH, which introduces CBA tech to boost performance and efficiency.

Kioxia is also preparing future flash memory generations using two methods. The first will add more layers for capacity, while the second blends new CMOS designs with older cell structures to keep investment costs in check.

You might also like

• These are the largest HDDs and SSDs on the market at the moment
• We've also rounded up the best workstations around today
• Another tech publisher has broken the Pi calculation world record