Feed aggregator

The Dell Tower Plus is one of those machines that impressed us during testing. It's such a versatile and configurable tower desktop that's suitable for everything from basic office use to full-blown creative workstation, whether out of the box or upgraded later down the line.

So, we were really pleased to see the Dell Tower Plus Desktop (EBT2250) is now $2000 (was $2530) at Dell. It's our top-ranked business computer for all professionals, with this performance-driven model equipped with a Core Ultra 7 CPU, 32GB DDR5 RAM, 1TB SSD, and RTX 5060 Ti GPU.

Should you buy it?

✅ Buy the Dell Tower Plus if...

You want one machine that covers both productivity and gaming without compromise. This configuration's RTX 5060 Ti and 20-core Ultra 7 chip give it real headroom for creative work, multitasking, and modern games alike, and the tower format leaves room to upgrade later.

❌ Skip the Dell Tower Plus if...

You want a compact or small-form-factor build, or care about aesthetics — our review flagged the minimalist, fairly plain design as one of its few weak points, along with the basic wired keyboard and mouse it ships with.

Intel Core Ultra 7 265 | 32GB DDR5 | 1TB SSD

This is our pick for best business PC overall - specifically because it's ultra-configurable and future-proof, whether you need a basic office machine or creative workstation. The deal we've selected has a Core Ultra 7 265 chip, 32GB DDR5, and 1TB SSD, with Nvidia GeForce RTX 5060 Ti graphics card. This feels like the best all-rounder, but you can easily switch out for the specs you need with discounts across the range. View Deal

Why we recommend it

In our 4.5-star review, we found the Dell Tower Plus is a machine that "does it all," praising how the same chassis scales from a barebones office machine up to a genuine creative or gaming rig, while staying quiet and easy to upgrade later.

What makes the Tower Plus line worth considering is just how configurable it is without changing the chassis or port layout. For that reason, we rank it the best business PC in our guide. Every professional can get something from this machine.

This particular build lands in a genuinely useful middle ground: enough GPU for 1440p gaming and GPU-accelerated creative work, enough CPU for heavy multitasking, and a straightforward path to upgrade RAM or storage down the line rather than being boxed in by a sealed or proprietary design.

Wi-Fi 7 connectivity, Windows 11 Home, and a wired keyboard and mouse round out this deal.

Price Context & Historical Value

It's usually tricky to track pricing history from Dell, but based on our previous coverage $1999.99 is the standard deal price for this desktop PC. Not an all-time low, but still a significant discount compared to full price.

You can check out all the EBT2250 configurations currently discounted at Dell by clicking here.

The Catch: What to know before you buy

Our review's main criticisms were about presentation rather than performance: the design leans minimalist to the point of looking plain, and the included keyboard and mouse are basic wired accessories rather than anything premium.

In our opinion, neither affects how the machine performs, but budget for your own peripherals if that matters to you.

For a desktop that flexes between work and play without complaint, this configuration is a strong buy at $1,999.99.

• See all desktop deals at Dell.

It's been a year since a gunman opened fire on the CDC and killed a police officer. RFK Jr. visited the CDC today to remember the victim.

The race between Susan Collins and Troy Jackson is one of the nation's most high-profile Senate contests this year -- and may ultimately be decided by an especially powerful voting bloc: Maine women.

• Hivemind completes first autonomous swarming mission on water in Taiwan
• Drone boats identified and escorted a vessel during a coordinated surveillance exercise
• SeaShark 600 and 800 vessels operated together under a single autonomous command

An autonomous software system called Hivemind has directed a group of unmanned boats through a live mission for the first time on open water.

Developed by Shield AI, the AI tool took control of two vessel types built by Thunder Tiger during a staged exercise off Pingtung, Taiwan.

Engineers designed the exercise to test whether one AI system could manage several drone boats acting together as a coordinated unit.

Autonomous software directs coordinated drone boat mission

Two vessel models, the SeaShark 600 and the SeaShark 800, carried out the reconnaissance work under Hivemind's direction during the exercise.

Rather than following pre-set routes, the software built its own path in real time using radar feeds, camera imagery, and ship-tracking signals pulled from the area.

It located the search zone on its own, spotted an object worth investigating, and then organized the fleet's response around that discovery.

Once the object was found, the boats worked together to guide a separate ship safely away from the area under watch.

The trial marked the first time Hivemind completed an on-water swarming exercise, extending technology previously associated with autonomous operations across different military platforms.

“Hivemind continues to prove that true mission autonomy can scale across domains, platforms, and functional assets, unlocking new capabilities for distributed intelligence and maritime security missions,” said Nathan Michael, chief technology officer at Shield AI.

“As we continue working with Thunder Tiger to expand the scale and sophistication of autonomous maritime teams, these systems will fundamentally reshape maritime surveillance operations and strengthen deterrence through resilient, intelligent operations.”

The companies began collaborating through a strategic agreement aimed at integrating Hivemind autonomy software across both maritime and airborne unmanned systems.

Companies plan broader maritime deployments

Shield AI and Thunder Tiger plan to expand their collaboration by integrating additional sensors and improving coordinated swarm capabilities across autonomous maritime platforms.

The companies also intend to conduct open-ocean deployments supporting future United States and allied military applications.

According to Thunder Tiger, the latest exercise also examined how fleets of unmanned surface vessels could improve maritime awareness by detecting, classifying and identifying contacts during security operations.

“By integrating AI pilots into our portfolio, we can enable distributed, networked teams that coordinate dynamically to build a more complete understanding of the maritime environment … while helping operators respond with greater speed, scale, and effectiveness,” said Gene Su, board director and general manager of Thunder Tiger Corp.

“This demonstration marks an important step … and we expect our partnership with Shield AI to continue delivering increasingly sophisticated autonomous systems that strengthen Taiwan's defense posture.”

The demonstration adds to ongoing efforts to advance autonomous maritime operations using coordinated unmanned surface vessels.

Future activities will likely focus on integrating additional sensors, expanding swarm capabilities, and conducting open-ocean deployments.

Via The DroneFront

GMKtec is behind some of the most capable mini PCs we've ever tested, and the K8 Plus is no exception. And I just found the GMKtec K8 Plus mini PC for $650 (was $900) at Amazon.

This configuration features the high-performance Ryzen 7 8845HS processor, paired with 16GB DDR5 RAM, and 1TB SSD. But the real highlight for us is the connectivity. Specifically, this mini PC comes with dual USB4 ports and an OCuLink port.

The 32GB/1TB K8 Plus is also on sale for $740 (was $750) direct from GMKtec, while the 32GB/512GB model is $689 (was $900) at Newegg.

Should you buy it?

✅ Buy the GMKtec K8 Plus if...

You want desktop-class performance and expandability in a footprint that fits behind a monitor. The upgradeable RAM and dual M.2 storage, plus an OCuLink port for a future external GPU, make this a genuinely future-proof pick for a dorm room, home office, or secondary workstation.

❌ Skip the GMKtec K8 Plus if...

You care about design details — our review noted the transparent twist-off top cheapens what's otherwise a clean, metal-cased build. You'll also need an external graphics card connected if you need serious dedicated graphics power, since the integrated Radeon 780M is fine for light gaming but not built for demanding titles alone.

AMD Ryzen 7 8845HS | 16GB DDR5 | 1TB SSD

This Windows 11 Pro mini PC from GMKtec comes equipped with a Ryzen 7 8845HS chip (8-core, 16-thread, up to 5.1GHz, 16MB L3 cache), 16GB of DDR5 RAM (expandable to 96GB) and a 1TB PCIe SSD. You'll also find Radeon 780M integrated graphics, with the option of hooking up an eGPU via the dual USB4 ports and an OCuLink port. We love the inclusion of dual 2.5G Ethernet, alongside HDMI 2.1, DisplayPort 2.1, Wi-Fi 6, and Bluetooth 5.2. View Deal

Why we recommend it

In our 4.5-star review, we called it an "understated mini PC that delivers far more power than you'd expect for the price," and found it handled demanding day-to-day tasks with total ease during testing, backed by a genuinely capable set of upgrade options.

The OCuLink port is the standout feature here. It gives this mini PC a genuine path to external GPU performance that most machines this size simply don't offer, which matters if your needs might grow beyond what the integrated Radeon 780M can handle.

Combined with dual 2.5G Ethernet and 16GB of upgradeable RAM, this machine is built to keep pace as your workload changes, not just handle what you need today.

Price Context & Historical Value

According to CamelCamelCamel, this is the lowest price I've seen on Amazon for the 16TB/1TB configuration. However, I've previously seen the 32GB/1TB version selling for $630 back in February - before the RAMpocalypse truly hit and we saw price risess across the board. Previously, the 32GB/2TB version was selling for $810 (down from $940), a deal that ended today.

Alternatively, GMKtec has the 32GB/1TB model for $740 (was $750), and Newegg is selling the 32GB/512GB model for $689 (was $900).

The Catch: What to know before you buy

Our main design complaint in the review was the transparent, twist-off top panel, which looks a bit cheaper than the rest of the metal chassis suggests it should. On the performance side, the integrated Radeon 780M graphics handle light gaming and everyday creative work well.

However, anyone chasing serious frame rates in demanding titles will want to rely on the OCuLink port for an external GPU rather than expect it from the iGPU alone.

For desktop-class performance in a genuinely small footprint, this is one of the better mini PC deals we've tracked. You can see more in our guide to the best mini PCs we've tested, and save with the best mini PC deals we've uncovered this week.

• See all mini PC deals at Amazon.

Trump is once again threatening to fire Cook over unproven allegations of mortgage fraud after an earlier effort to remove Cook was struck down by the Supreme Court.

(Image credit: Kevin Dietsch)

• A developer has a 28.9-million-parameter model generating TinyStories-style text at 9.88 tokens per second, fully offline, on a $10 microcontroller
• It was achieved by fitting a language model on a chip with 512KB of RAM by leaving most of it in flash storage
• The project is available on GitHub under an MIT license

A developer going by slvDev has a 28.9-million-parameter language model generating text on an ESP32-S3, a microcontroller built for sensor nodes and smart plugs, at 9.88 tokens per second, with nothing leaving the chip.

The project, esp32-ai, went up on GitHub under an MIT license in late July 2026, was showcased on the Better Stack YouTube channel, and has since collected over 3,600 stars and has more than 470 forks of the underlying code.

Given the underlying hardware's limitations, the attempt meant the model, which was trained on the TinyStories dataset by Microsoft Research, had to be scaled down.

Still a capable LLM despite the loss in accuracy

The arithmetic involved here is brutally unforgiving; weights typically need one to four bytes per parameter just to sit in memory, so 28.9 million parameters at 16-bit precision still require nearly 60 MB of RAM. An ESP32-S3 offers 512 KB of SRAM and 8 MB of PSRAM that is woefully inadequate at first glance.

Quantization closes the first part of the gap: dropping the weights to four-bit precision trades a little accuracy for a 75% reduction, bringing 60MB down to 14.9MB.

The second step is where the project earns most of its attention: most of a language model's parameters sit in embedding tables that the model reads from rather than computes against, and something you only read can live somewhere slow.

Borrowing Per-Layer Embeddings from Google's Gemma 3n, slvDev moved about 25 million parameters (roughly 12 MB at four-bit) into the board's 16 MB of flash, pulling around 6 rows, about 450 bytes, per token.

What remains in working memory is closer to 2 MB: the dense core and output head in PSRAM, and activations and norm weights in SRAM. Each tier stores whatever is read at its own frequency, and the chip's memory hierarchy serves as the architecture that enables the LLM.

Offloading weights to storage is not new; it is how people coax frontier-class models onto hardware that has no business running them, but it usually destroys throughput, turning tokens per second into seconds or minutes per token.

The PLE approach avoids that because the offloaded weights are accessed sparingly, so flash bandwidth never becomes the bottleneck. The resulting performance, 9.88 tokens per second, is faster than most people read.

The LLM has its limitations, however: the model writes short, mostly coherent stories and will not answer questions, follow instructions, write code, or know facts. The developer says the limit stems from the small part of the model that performs the reasoning (~4M parameters) and that the memory trick does not change its capabilities. His repository also ships a second model, Barista, which expressly answers espresso questions only (pun intended).

In short, the approach doesn't make small models smarter, but it does allow them to offload data within the board onto chips they could not previously run on, making it an interesting choice, though one with limited practical uses in its current iteration.

If you shoot video, edit photos, or just want a fast backup drive that can survive being tossed in a bag, this deal is worth grabbing. The SanDisk Extreme 1TB Portable SSD is $227 (was $256) at Amazon.

This is SanDisk's newer Extreme Portable SSD rated for read and write speeds of up to 2000 MB/s over USB-C - a notable jump over the 1050 MB/s the excellent previous generation of this drive topped out at. It's rated IP65 for water and dust resistance and backed by a 5-year limited warranty.

Should you buy it?

✅ Buy the SanDisk Extreme 1TB if...

You need a fast external drive for video editing, large file transfers, or backing up a laptop, and want something rugged enough to travel with. The 2000 MB/s and IP65 ratings make it well-suited to fieldwork, whether that's a shoot, a shared desk, or a backpack.

❌ Skip the SanDisk Extreme 1TB if...

Your device or cable setup can't take advantage of the faster USB 3.2 Gen 2x2 interface — you won't see the full 2000MB/s speeds on an older USB-C port, and a cheaper standard SSD would serve just as well.

A 1TB portable SSD rated for up to 2000MB/s read and write speeds over USB-C (USB 3.2 Gen 2x2), with IP65 water and dust resistance, 256-bit AES encryption with password protection, and a 5-year limited warranty.View Deal

Why we recommend it

SanDisk's Extreme line has always impressed us - we awarded the older model 4.5 stars in our review, with a TechRadar Pro Recommends badge. At the time, we called it "hands-down one of the best external SSDs (rugged or not) on the market, and its 1TB version is particularly attractive."

This is a faster version, but it's still the same design, including the durability rating for outdoor use (or a particularly violent commute).

The jump to 2000MB/s puts this drive well ahead of the standard 1050MB/s tier most portable SSDs in this price range are still stuck at, which matters if you're regularly moving large video files or full camera card dumps. Pairing that speed with IP65 durability and a 5-year warranty is still fairly rare below $250.

Price Context & Historical Value

Since I first covered this deal yesterday, I've seen prices fluctuate between $240 (was $300) at Best Buy to an all-time low of $215 at Amazon. The highest price I've seen is $270. So, at $227 it's not the cheapest but still the best deal around right now.

The Catch: What to know before you buy

To actually hit the rated 2000MB/s speeds, you'll need a USB 3.2 Gen 2x2-capable port on your computer — plug it into an older USB-C or USB-A port, and you'll see noticeably lower real-world speeds. It's also worth knowing that IP65 resistance protects against splashes and dust, not full submersion, so this isn't a drive to take diving.

• See all SSD deals at Best Buy.

Welcome to the Blu-ray Bounty July 2026. This is where we test the latest 4K releases of each month that we think will be perfect for showing off your home theater. If you’re new to the Blu-ray Bounty, you can check out previous editions here, where we’ve reviewed over 100 discs since starting in November 2024.

4K Blu-ray is easily one of the best ways to enjoy a movie at home, especially when it’s on the best TVs and best soundbars. I personally use it as my main source for testing AV equipment, thanks to its uncompressed picture and audio quality.

In last month’s June 2026 Blu-ray Bounty, I tested seven discs, including Speed Racer (which is one of the best discs I’ve ever tested, and you can expect to see it as reference source in future reviews), EPiC: Elvis Presley In Concert, Bullet In The Head (1991), Body Heat, Avatar: Fire & Ash, Five Easy Pieces and 36 Hours.

In this month’s list, I’ll be testing 4 discs, and I’ll be using our usual setup. This consists of the LG G6, one of 2026’s best OLED TVs, the Panasonic DP-UB820, the best 4K Blu-ray player for most users on the market, and the Samsung HW-Q990C, the brand’s flagship 11.1.4 channel Dolby Atmos system from 2023.

As always, I’ll only be discussing the video and audio quality of each disc, not the movie itself.

The Super Mario Galaxy Movie (Universal Pictures) Nintendo / Universal Pictures / Future Nintendo / Universal Pictures / Future Future

The Super Mario Galaxy Movie, sequel to 2023's The Super Mario Bros. Movie, follows Mario, Luigi, Peach Yoshi and Toad's adventure to rescue Princess Rosalina, who's been kidnapped by Bowser Jr. The film stars Chris Pratt, Anya-Taylor Joy, Charlie Day and Jack Black, who reprises his role as Bowser.

The Super Mario Galaxy Movie looks gorgeous on 4K Blu-ray. This disc is so impressive, that I've even started using it for reference testing, and you can read my full thoughts on it here. Below is a summary about the disc's picture and audio.

It's all about color reproduction here, which is bold and vibrant, generating some amazing pop on screen. As the Lumas listen to Rosalina's bedtime story, the color of each Luma is punchy with some amazing highlights, especially with Dolby Vision HDR. The animation throughout is crisp, delivering 3D-like textures and depth that is seriously impressive.

If you have a bright display at home, you'll be rewarded with some beautiful brightness too. The star that one of the Lumas turns into dazzles on screen, and anytime Rosalina uses magic, there are stunning, glittery colors that illuminate the screen. The movie delivers excellent contrast too, particularly in space scenes where the dark blue sky of space balances nicely with the bright stars and vivid characters.

The Super Mario Galaxy Movie has an impressive Dolby Atmos soundtrack included too. There's excellent detail and precision here, as Peach and Toad first fly through the space on the star, the twinkling trail behind the star follows its path accurately, and is well defined in the mix, clearly audible even in the uplifting score. As characters fly towards Bowser Jr's ship, height effects are fully utilized, with the character's flight mapped overhead to create a fully immersive sound. There's even great demonstrations of bass throughout, as character's fire up ship engines or crash through things, generating plenty of impact.

To Live and Die in L.A (Arrow Video)Arrow Video / Future Arrow Video / Future Future

To Live and Die in L.A. follows the story of two Secret Service agents, Richard Chance (William Petersen) and John Vukovich (John Pankow), who are in pursuit of counterfeiter-turned-killer Eric Masters (Willem Defoe) after Masters kills Chance’s former partner.

To Live and Die in L.A. looks excellent in 4K in this restoration, presented by Arrow Video. Textures have been cleaned up well, with skin looking realistic, and more intricate details such as scars and marks on people’s faces delivered with real clarity in close-up shots. Colors have seen beautiful pop, with the Dolby Vision grade really adding some serious punch. Reds, such as those on the US flags throughout, in particular jump off the screen.

There’s some strong contrast throughout, as well as some refined shadow detail. In dimly-lit areas, overhead lights create strong shadows across people’s faces, with a stakeout scene with Chance and Vukovich acting as one of the best demonstrations of this. Black tones are deep throughout as well, in both environments and people’s clothing.

There are two soundtracks included here: the original stereo audio and a DTS-HD 5.1 MA mix. Using the 5.1 MA mix, there’s some seriously good detail here. As Masters makes counterfeit notes, the whirring of the printing press balances nicely with Wang Chung’s score. Even intricate effects in the scene, such as Masters cutting out the notes and washing the surface, are clearly audible. During a night-time stakeout, the pouring rain is accurately mapped to the rear channels, even creating a sense of height. Speech is clear throughout, and the soundstage is wide enough to give all elements room to breathe.

Alice In Wonderland (1951) (Disney)Disney / Future Disney / Future Future

Alice In Wonderland, based on both Alice’s Adventures in Wonderland and Alice Through the Looking Glass by Lewis Carroll, tells the story of Alice, a young girl who ends up in a magical world where she meets a weird and wonderful cast of characters, including the evil Queen of Hearts who rules over Wonderland.

Alice in Wonderland looks great in 4K. Colors are bold and accurate in equal measure, faithfully rendering that classic Disney look. While most colors are beautifully vivid, it’s reds and pinks that really stand out on screen, looking particularly dazzling. The animation style here looks refined and has been nicely upscaled, giving characters more depth.

On the LG G6 I used, bright colors such as the white of Alice’s clothes and the gold of the doorknob have striking brightness and vibrancy to them, aided by the Dolby Vision HDR featured on this disc. While I usually use Filmmaker Mode on the G6, I found this movie did benefit from the brighter Cinema Home mode for some added punch.

While there is still some film grain present for cinephiles, the movie has been refined quite a lot and this does give the animation a very clean look, making it look much more contemporary than its original 1951 animation style would suggest.

There are two soundtrack options available: DTS-HD MA 5.1 and 2.0 mono. The 2.0 mono is more faithful to the original, but the 5.1 MA mix has a nice balance to it, giving every element depth and warmth, while letting them all breathe. The score is mapped to the rear speakers, allowing the front channels to deliver speech and other effects. Outside of the score and vocals, there aren’t many other elements to show off your home theater, but this 5.1 mix does sound brilliant.

Hairspray (1988) (Criterion)The Criterion Collection / Future The Criterion Collection / Future Future

Set in 1962, Hairspray follows Tracy Turnblad (Ricki Lake), a teenager who achieves stardom on a local dance show, and attempts to use her newfound fame to advocate for more racial integration on the dance show. Directed by John Waters, the movie also stars Divine, Sony Bono, Jerry Stiller and Debbie Harry.

As you’d expect from a movie set in the 1960s, there’s some seriously bold visuals here. From the bright pinks of the Turnblad home to the striking red of Link’s car, colors have some really nice pop on screen. Textures have been upscaled effectively, giving them a more 3D-like look, though not quite to the extent from other Criterion restorations such as 3:10 To Yuma. Finer details have also been refined, with individual strands of hair given some nice depth and volume.

There’s a lot of grain still present here, so cinephiles will be happy. Black tones are deep as well, shown predominantly in people’s hair. There’s some nice contrast too, as shaded areas contrast with the bright lights of the studios for the dance shows themselves.

For audio, there are two soundtracks available: DTS-HD MA 2.0 surround and DTS-HD MA 5.1. The 5.1 soundtrack does a great job of giving the soundtrack room to breathe, with a good balance between the speech and stacked tracklist of songs used for the dance scenes. More intricate effects such as the sound of shuffling feet and clapping hands during dance scenes have a nice clarity, clearly audible over the much louder songs.

The company’s new “2.0” plans and 36-month installment plans provided cover for removing what was billed as a limited-time offer.

A federal appeals court says above-ground construction cannot continue without approval from Congress, siding with historic preservation advocates.

(Image credit: Tierney L. Cross)

• Researchers tested AI-generated patches on six CVEs with poor success rates
• Many fixes failed, altered behavior, or introduced new vulnerabilities
• Guidance improved outcomes, leading to FLAWED evaluation harness release

When using Generative Artificial Intelligence (GenAI) to fix vulnerabilities, security professionals are most of the time just robbing Peter to pay Paul, experts have warned.

Researchers from 1Passwords Off-by-1 Labs analyzed fixes proposed by two frontier models - ChatGPT 5.5 at “medium” effort, and Claude Opus 4.8 at “high” effort.

As an experiment, the researchers took six recently disclosed CVEs and produced 6,080 patches using two frontier, cyber-capable reasoning models. The results were underwhelming to say the least - of all the proposed patches, just a quarter (26%) fully resolved the issue.

FLAWED work?

This obviously leaves plenty to be desired, as half (49.3%) of the patches failed to fix at least one existing exploit path. A fifth (20.1%) fixed the original issue but changed application behavior, while 2.3% introduced new security issues. Funny enough, 2.2% failed to fix the vulnerability while also introducing additional exploit paths, as well.

Even among the patches that might be considered (26% of clean ones and 20.1% of those that changed app behavior), more than a third were fragile and not entirely addressing the underlying problem.

The researchers created an acronym for automated LLM patches: FLAWED (Fix-Like Artifacts With Embedded Defects), and warned against letting AI work without human oversight: "The expected value of a fully LLM-generated, non-human-reviewed patch is a net-negative by a considerable margin."

Results drastically improved when the AI was given better context, the researchers further explained. Before working on any patch, human developers are usually given initial guidance. When AI is given proper guidance, its success rate rises to 65%. Incorrect guidance, on the other hand, drops the success rate down to 15.2%. The difference between humans and AI is that humans are better at catching misleading information and poor guidance.

This doesn’t mean developers will, or should, abandon AI. Worst case scenario is that developers will spend more time reviewing AI-generated fixes which could increase cognitive load and still end up being net negative. Therefore, the researchers released a patch evaluation harness called FLAWED, which organizations can now use to determine the effectiveness of their AI-generated fixes.

Via The Register

The ChatGPT maker is reportedly working on a small, portable AI smart speaker, but it better not look too much like Apple’s designs.

The internationally acclaimed horseman defied the conventional method of breaking horses and developed his own gentler approach — based on horses' non-verbal language. Roberts spoke with Fresh Air in 1997.

Our critical utility infrastructure can make the same classic mistakes as we do with our everyday connected devices.

The bill is meant to stifle Russia's economy by allowing for tough new tariffs against major importers of Russian energy. The legislation was led by Lindsey Graham before his unexpected death in July.

(Image credit: Heather Diehl)

• Attackers cloned AI skills, later adding malicious code to steal credentials
• Zenity Labs found millions of installs and dozens of dangerous skill variants
• Vercel and Microsoft removed malicious skills, but manual removal is still required

AI skills, instructions that teach AI agents how to do certain tasks and thus extend their capabilities, are increasingly being used in supply chain attacks, researchers have found.

Security experts at Zenity Labs uncovered a credential-stealing campaign on skills.sh, a public registry (essentially an app store) for AI agent skills. In the registry, belonging to Vercel (a cloud platform for web applications), threat actors were cloning existing skills, creating typosquatted lookalikes which, at first, did nothing malicious.

However, after a little time had past, and the skills amassed a solid download count, the attackers introduced malicious code instructing the AI agents to, among other things, exfiltrate SSH keys, cloud credentials, Git and package manager tokens, Kubernetes and Docker configurations, database credentials, infrastructure-as-code credentials, environment files and service account files. The agents were then told to package the stolen information with host metadata and send it to the attackers.

Dozens of malicious skills

While Zenity Labs could not say exactly how many people fell victim to this attack, they did stress that a single skill family amassed more than 1.7 million aggregate installs (not unique users).

And that is just one skill family, in a sea of malicious skills. The researchers also said they found “dozens” of additional skills exhibiting either malicious or dangerous behavior. Almost a third (30%) of identified dangerous skills abused Claude Code and OpenClaw to drop malware to their targets, as well. Also, Zenity found “hundreds” of reserved and empty package names that were being kept for future attacks.

These findings show how quickly cybercriminals adapt, and how creative they can get when it comes to abusing new tech. In essence, this campaign is an AI spin on a software supply-chain attack, being similar in spirit to incidents where attackers compromise an existing trusted package or repository, and later push a malicious update.

Following responsible disclosure, Vercel and Microsoft removed the identified skills, but Zenity warns that those who installed them before won’t be safe until they remove them from their systems manually.

It's a smart donut. That's the only conclusion I can draw about OpenAI's first, groundbreaking piece of AI hardware after reading Bloomberg's revealing but unconfirmed report.

Here's the TLDR summary of these rumors: This donut-shaped, hockey puck-sized AI companion, designed with Jony Ive's LoveFrom studio, can sit on a table, be carried (or maybe worn). It'll be festooned with sensors, cameras, microphones, and speakers. Meanwhile, it'll be a vessel to bring ChatGPT closer to you and your life. It'll even shape-shift a bit to indicate a response (can a donut shrug?).

OpenAI might, Bloomberg claims, price it between $300 and $400 (or around £220-£300 / AU$425-AU$570). I know. That's almost instantly a hard no for an AI companion that has no defined purpose besides offering always-with-you AI. Even the woebegone Rabbit R1, which also features a camera, speakers, some sensors, and even a rotating camera, costs just $199 (or about £150 / AU$280).

After reading through this, I still wasn't feeling it. Why do we need this gadget? What problem does it solve? Really, why do we need any AI on or near our bodies?

If this product arrived three years ago, would people, including me, feel differently? After all, in 2023, we were all excited about the potential of generative AI. These days, it's fair to say half of us hate it or at least deeply distrust it and hate what it's doing to our resources (water and energy) and landscapes (data centers).

On the other hand, maybe I was judging OpenAI's upcoming device too harshly, especially without having even seen it. That sparked an idea.

Bloomberg's report included enough detail to create an image in my mind's eye, and I wondered if the leading AI platforms, Gemini and, yes, ChatGPT, could use a prompt to generate simulacra of the devices, in situ.

Gemini and ChatGPT take a run at it

I started with this prompt:

"I need an image of a consumer electronics gadget that is shaped like a donut and the size of a hockey puck. It should include tiny microphone holes and grills for a pair of speakers. It should have moving parts, maybe pieces that shift against each other without massively deforming the donut shape. Let's make a pair of them in white and gray on a table. Let's also include an image of someone wearing one like a pendant."

And in full disclosure, after getting my first couple of images, I added this prompt because I realized I left out a pair of key features:

"Without changing the design, can you add a camera and a couple of small black sensors? I want the image to otherwise remain unchanged."

These AI image generation systems are now good enough that they can maintain consistency between image generations (which they did here), so I'll only show you the finished products from both platforms.

Here's the final render from Gemini Pro:

(Image credit: Gemini)

At a glance, they're kind of cute, looking like giant, digital Lifesaver candies. The speaker grills are kind of huge, but I like the subtle placement of the camera and sensors. I can infer the possibility of movement from the multi-part frames.

Gemini is, I can see, a bit unsure if this is a wired or wireless device, but it has made at least one intriguing leap. It assumes, for instance, a connection to a phone-based platform called Aether (the gadgets may even be called "Aether"). In medieval times, according to Gemini, Aether was considered the fifth element. "It was believed to be the pure, heavenly material that filled the region of the universe beyond the terrestrial sphere," wrote Gemini. Yes, that feels a bit like the proliferation of AI.

The fictional device does look as silly as a pendant as I expected, but overall, I'm more intrigued by this device than I was before seeing the Gemini render.

Now let's take a look at the ChatGPT (Free) render:

(Image credit: ChatGPT)

This is a simpler, even plainer device and might fit in a bit better with Jony Ive's previous design oeuvre. I like the low-key speaker grill design, and don't mind the smaller activity lights.

I appreciate that ChatGPT helpfully includes a schematic that breaks down all the key features and even offers a look under the hood. It also manages to look a little less obtrusive as a pendant.

Let's get real

Obviously, this is all guesswork by Gemini and ChatGPT, and it's based on prompts built on rumors. Put another way, OpenAI's upcoming AI wearable could end up being significantly different than either of these generative images.

Somehow, I don't think so. I have a feeling that we will see something small, round, lightweight, and very, very intelligent. The launch, when it happens, will be exciting and buzzworthy. Ive's voice will likely drive the brand narrative as the donut-shaped device floats in an all-white background before settling on a desk or, better yet, in the palm of someone's hand.

That moment, though, won't change the trajectory of OpenAI's AI gadget. When it arrives, it will face an uphill battle to attract consumers who are already suspicious of AI's growing influence on the world. They're tired of AI slop, angry about resources and jobs, and the last thing many want is an AI companion reminding them daily about their frustration.

I don't even know what to say about a donut... er... gadget that squirms in your hand. That sounds positively creepy.

If there's any bright side to these rumors, it's that they probably put to bed the idea that OpenAI is allegedly stealing Apple product secrets. The iPhone maker would never build something like this.

I could be wrong about OpenAI's chances in this space. This might be the first AI gadget to break through. I mean, look at those designs. Aren't they cool? Sure, they are, but they're not reality, and we all know that this gadget will soon face a very harsh reality.

A good ergonomic office chair can make spending long hours at your desk considerably more comfortable, but some of the best models can cost a small fortune. Thankfully, I've found a terrific deal on FlexiSpot's C7 office chair for $300 (was $430).

Better still, this is a chair we highly recommend and have ranked as our top-rated mid-range ergonomic office chair in our buying guide. In the UK, the C7 is priced at £259.99 (was £379.99).

The C7 combines a breathable mesh design with self-adaptive dynamic lumbar support, adjustable seat depth, 3D armrests, and a 4D headrest, giving you plenty of ways to tailor the chair to your body and working position. It also reclines from 90 to 130 degrees, supports up to 300 pounds, and comes with a reassuring 10-year warranty.

Should you buy it?

✅ Buy the FlexiSpot C7 if...

You spend long hours working at a desk and want plenty of ergonomic adjustment without moving into the much higher price brackets occupied by premium office chairs. Its adjustable seat depth, lumbar support, armrests, and headrest make it easier to find a comfortable position, while the 300-pound capacity allows it to accommodate most people.

❌ Skip the FlexiSpot C7 if...

You want a simpler chair with fewer adjustments, or you're shopping on a tighter budget. The C7's ergonomic features are aimed primarily at people who spend long periods at a desk, so occasional home-office use may not justify spending out.

The FlexiSpot C7 supports up to 300 pounds and features dynamic lumbar support, adjustable seat depth, 3D armrests, a 4D headrest, 90-to-130-degree recline, and a breathable mesh seat.

In the UK: now £260 (was £380)View Deal

Why we recommend it

In his glowing 4.5-star review our expert Collin said: "This chair is an excellent solution for office managers looking for a quality ergonomic office chair for their employees; it's great for home office use, for your office, cubicle work, and more."

There's an impressive amount of adjustment available for the money, particularly the dynamic lumbar support and adjustable seat depth.

The mesh construction should also help with airflow during long working days, and the 10-year warranty provides some welcome reassurance if you're buying a chair for everyday use.

Price context & historical value

At $299.99, the C7 costs $130 less than its usual $429.99 price, equivalent to a discount of just over 30%. That brings a chair we already consider a fantastic mid-range choice much closer to the price of less sophisticated ergonomic models.

Given the combination of adjustability, 300-pound capacity, lengthy warranty, and the current price, this would make a fantastic home office upgrade and is an easy recommendation from me.

FlexiSpot offer regular sales throughout the year, and by and large, this is the standard deal price. I can't remember seeing it cheaper (although I've certainly seen it more expensive).

The Catch: What to know before you buy

This configuration doesn't include a footrest. If you want one, adding the feature increases the price to $319.99, which isn't that huge a leap.

The mesh construction won't suit everyone, particularly if you prefer a softer, cushioned seat, but you can switch to foam for the same price (although not with a footrest).

FlexiSpot also recommends the C7 for people between 5 feet 5 inches and 6 feet 11 inches tall. Our reviewer Collin stands a little over six foot and had no issues using the C7, but smaller users will want to pay particular attention to the adjustment ranges for maximum comfort.

• US: See all deals at FlexiSpot
• UK: See the full sale at FlexiSpot.co.uk

He’ll be among the many celebrities attending the launch party for the new Pixel devices.