Technology

• Phishing campaign used fake Teams notifications to route victims to a genuine Microsoft sign-in page
• Rather than stealing passwords, attackers asked victims to approve permissions for an attacker-controlled app, gaining access to mail, files, Teams, SharePoint, OneDrive and calendars without defeating MFA
• Check Point says the technique has been commoditized in 2026 into a rentable service; the practical defense is restricting app consent rather than relying on users to spot a fake

A phishing campaign that ran from late June into July 2026 did something that breaks most of the advice organizations have spent a decade teaching their staff: it sent victims to a real Microsoft login page.

Check Point's email research team, which disclosed the campaign, identified more than 200 phishing emails targeting users across roughly 120 organizations worldwide.

The lure was a fake Microsoft Teams notification with a genuine destination; what actually compromised accounts was not a stolen password but a permissions prompt that the victim clicked through voluntarily.

A sophisticated attack that relied on tricking users into granting permissions

Check Point noted in its brief that what actually compromised the accounts was not a stolen password but a permissions prompt that the victim clicked through voluntarily. This is a reminder of a stark change in attackers' tactics: they have stopped forging Microsoft's front door and started walking through it.

The email appeared to be a Microsoft Planner task-assignment notification. The sender name read "There's New Activity On Team," the subject line claimed that HR had sent three messages via Teams chat, and the body referenced a payroll and benefits update, along with a counter showing four overdue employee tasks.

To someone who works in security, the message had its own telltale signs of being a typical phishing attempt: every link in the email, including both call-to-action buttons, routed through the same redirect. And the visible sender address belonged to the recipient's own organization, meaning the email appeared to have been sent to the same person it came from.

The link opened a real OAuth authorization URL on login.microsoftonline.com, not a look-alike domain. Signing in displayed a permissions prompt asking the user to approve the permissions or accept them on behalf of their organization.

If they did, Microsoft redirected the browser to the redirect address specified in the original request, which in this particular campaign was an AWS API Gateway endpoint under the attackers' control. The authorization code was delivered there, and the attackers exchanged it for access. No password was stolen at any point, and there was no fake page to spot.

This is not unlike how the phishing-as-a-service Kali365 platform compromises Microsoft accounts, but instead of stealing session cookies or OAuth tokens, it opts for a more permanent illicit grant of consent.

This runs counter to the usual security training checklist, which emphasizes adhering to norms rather than going against the grain; users are told to check the URL, look for the padlock, and watch for misspelled domains. None of those measures matter because there is nothing forged to catch. The domain and certificate are Microsoft's, while the sign-in page is the one the user sees every morning, offering a false sense of security to a user not looking for this particular attack vector.

Multi-factor authentication does not help either; it protects the login sequence but not access to the user's data post-login. The attacker never needs the password or the second factor because they walk away with a token granted by the user's valid session, a technique called 'consent phishing'.

There are many ways to prevent this, but the simplest two are asking users to check every single permission/consent screen they click (the phishing attempt still requires users to allow it) and limiting access to permissions for user accounts that applications can request via Microsoft Entra at the system administrator level.

It would be prudent to do the latter at a minimum, even as Check Point notes that the campaign is no longer active because the underlying technique it used is not going anywhere.

• New study puts average cybercrime victim losses at $9,468, making a global annual toll at just over $1.24 trillion across 130.9 million victims
• Cybercrime is still massively underreported by both victims and authorities, so official numbers might be 'softer' than the underlying problem
• Some of the countries published limited, if any, financial data regarding cybercrime, making losses an estimate at best

New figures have claimed cybercrime victims lose $9,468 in the average incident, showing the scale of an increasingly global problem.

The report from Comparitech claims 130.9 million people are hit each year, and that the annual global toll comes to just over $1.24 trillion.

A December 2023 study from the company put those figures at $8,069 per victim, 88.5 million people and $714 billion in total losses respectively, highlighting a growing trend that sees a mix of illegal activity moving the needle further.

A growing problem with regional caveats

The United States remains a favorite for cybercriminals, topping the charts with 6.7 million victims losing $138.9 billion, a per-victim loss amount of ~$20,731, more than twice that of the global average.

This is in stark contrast with the next four countries on the list (Spain, France, Sweden and Turkey), all of which offered an average of approximately $10,000 per victim.

Russia, coming in 6th, reports a much larger number of victims than the four countries ahead of it, but offers a much lower per-victim loss estimate of $3,659. Interestingly, the total number of cybercrimes committed in 2025 dropped to 663,000 from 775,000 in 2024, despite an ongoing conflict with Ukraine, which often sees cyberattacks at both the industrial and localized levels by both parties.

However, a weakening economic situation, as well as Russia localizing many of its communication applications and restricting banking, might also mean that Russia's figures also remain inadvertently capped by policy decisions the country has taken.

With 18.8 million victims, India has the highest number of scam victims worldwide, even though its pro-rata number is considerably lower than the mean at ~$835 per victim, which may be attributable to the country's lower GDP per capita.

Interestingly, China, with 1.2 million victims, a fraction of its neighbor, managed to lose approximately $11.5 billion, a pro-rata number of approximately $9583, in line with global estimates by Comparitech and possibly fueled by the country's heavy-handed approach to cybercriminals, which saw it apply increasing amounts of pressure on neighboring Myanmar that culminated in it convicting and executing scammers arrested across the border.

Comparitech's $1.24 trillion figure is conservative, and the study acknowledges this, noting that it covers only victim losses. It notes that experts anticipated the global cost reaching $10.5 trillion in 2025 and calls its own $1.24 trillion a drop in the ocean by comparison.

That $10.5 trillion comes from Cybersecurity Ventures' 2016 report, which estimated $3 trillion for 2015 and projected it forward at an assumed 15 percent compound annual growth rate, describing the result as the greatest transfer of economic wealth in history.

At a time when AI automation offers better security, often allowing users to screen calls or leverage security applications that adapt on a case-by-case basis, the inverse is also true with hackers and cybercriminals considerably upping their game when it comes to bypassing security altogether; Comparitech's figures provide a sobering reality: if the industry estimates it has hold and are compared to the GDP of entire countries, the firm says it would rank 20th in the world in those terms alone.

With AI advancing in capability and availability, many scientists and experts have begun highlighting the risks of deploying this emerging technology. There's an overriding desire to deploy AI ethically, or to instill models with human-based values, but it remains to be seen how to standardize such an approach or even whether to regulate the industry. That's where organizations like the Stanford Institute for Human-Centered AI (HAI) come in.

Computer says no

When interviewed for the journal Issues in Science and Technology, the so-called 'Godmother of AI' Fei-Fei Li tried to explain how scientists and engineers can build AI that truly adheres to our values.

Quote of the day

This article is part of TechRadar Pro's QOTD project to provide an insight into the minds of the brightest and most recognized figures in the technology industry today and in years gone by. Read the full series here.

Her thesis around the responsible use of AI centers not on anthropomorphizing AI or considering it an intelligence capable of navigating the world and arriving at its own independent conclusions. Rather, it will reflect the values that we instill into it, whether that's through training or continued use of these tools.

She added in this interview that it's also about recognizing that AI, as any tool, can both empower and harm humans.

Values-driven AI

The organization that the American computer scientist leads, HAI, is an interdisciplinary research center that focuses on how best to guide AI development so that it benefits humanity and doesn't incur or lead to harm.

In recent years, there's also been an increased emphasis on whether to regulate AI nationally (or even globally) and what shape that might take. Doing so, however, is far easier said than done, with progress occurring at different paces, in different guises, and inconsistently across oceans.

There are, for example, different values that different nations or groups of people may wish to instill into the AI models they build that clash with the values of others.

• The latest Steam survey shows GPUs with 16GB are the most popular
• They're on 25.9% and have overtaken 8GB GPUs which are just behind on 25.32%
• There's been quite a shift towards 16GB boards over the past year, but with news of more price hikes incoming, upgrading to this loadout could be a painful process for gamers

Graphics cards with 16GB of memory are the most popular GPU among Steam gamers now, although the way GPU prices are going means that people aiming for a board packing plenty of video RAM (VRAM) are looking at a rather depressing future.

TweakTown spotted that the latest Steam hardware survey for July shows that 16GB of VRAM is the most prevalent memory configuration at 25.9%, up 1.4% over last month. That means it just dethrones the previous top of the VRAM table, which was 8GB, a loadout that slipped slightly from last month down to 25.32%.

There's not much in it, of course, but nonetheless, it's a telling stat (bearing in mind the usual caveats about variance in Valve's hardware report based on how the survey is distributed).

Of course, the majority of Steam gamers are still running with less than 16GB, and only 36% have a GPU with 16GB or more. So, almost two-thirds are running less than 16GB, and just under half (47%) have a graphics card with 8GB or less.

But it's a hefty rise in the count of 16GB GPUs when you consider that less than a year ago, in August 2025, just 6.8% of Steam gamers had boards with that VRAM loadout (based on a Notebookcheck report from the time). Compared to July 2026, that's nearly quadruple the number of 16GB-equipped rigs, with a close to 20% increase — pretty eye-opening.

Analysis: 16GB ambitions and the ever-worsening RAM crisis

(Image credit: Future / John Loeffler)

It seems that, at least based on Steam's stats, RTX 5060 Ti 16GB and RX 9060 XT 16GB graphics cards (and 9070 models) must have been selling well — and of course higher-end Nvidia cards. The RTX 5070 is doing well for itself, too, and is now in third position in Steam's GPU ranking (although that's a 12GB offering and it should really be 16GB).

The ranks of the 16GB GPUs on Steam will also be bolstered by older models, of course, from past generations. Also, the visibility of AMD Radeon graphics cards with 16GB of VRAM was improved by Valve earlier this year, when an issue was fixed whereby these cards were misreported (not detected and lumped in a generic Radeon category), so that will have boosted numbers too.

At any rate, the fact that more gamers are looking at future-proofing their GPU purchases only makes sense, as 8GB in particular is looking too lean at this point. But there's a problem with GPUs that have heavier VRAM configurations, and this is, of course, that the memory crisis hits these models hardest.

We've already seen that GPU price hikes are making themselves felt, and just a couple of weeks back, Nvidia seemingly told its card-making partners that the price it charges for memory supplied with its graphics chips is going up (for both GDDR6 and GDDR7). A fresh report from Tom's Hardware also talks about price hikes for RTX 5000 models in South Korea beginning this month, driven not just by the increased cost of video RAM, but also the GPU chips themselves.

We're told the upshot is that Nvidia's Blackwell GPUs could see price increases of up to 30% in the country, and similar hikes may happen elsewhere. AMD price hikes are expected for its Radeon line-up, too, and Team Red has already warned us that the second half of 2026 is going to be a tough one for PC gamers.

With those gamers craving 16GB of video RAM more often, and prices heading inexorably upwards — and the memory crisis expected to worsen as this year rolls on, with the misery to continue into 2027 by all accounts — it's a gloomy picture for beefier graphics card upgrades.

Beelink EQi Wildcat Lake: 30-second review

The Beelink EQi is essentially an ideal office machine. There’s nothing fancy about the design; in fact, it feels a little lightweight and plasticy; however, the balance of features is far greater than the specifications would initially have you believe.

As ever, this Mini PC is compact with its 126mm-square footprint that enables easy positioning around the desk; however, while the looks may not highlight anything special, it packs in an unusual amount of connectivity for a Mini PC at this price and with this level of CPU.

The major surprise here is that on the back, alongside the usual 2.5GbE LAN port, is a second, ultra-fast 10GbE port, which is the sort of specification you normally find on hardware costing far more, but also ideal if you’re connecting into a UGREEN IDX6011 NAS over a 10GbE network for archiving. Next to this network port are two Thunderbolt 4 ports at 40 Gbps, HDMI 2.0, and triple-display support, meaning that for integration into an office or studio this machine is highly tuned.

It’s also very neat with the dark grey ABS enclosure feeling cheap compared to aluminium rivals. Still, the 85W power supply is built into the chassis, so the only cable at the back is a mains lead, and those going to monitors and accessories, there’s no power brick to get in the way.

Open the base and the interior is exceptionally well organised, with both M.2 slots covered by a substantial metal heatsink that lifts off with two screws. The rear port layout is the odd decision, with the two Type-A ports placed on either side of the HDMI output rather than together, which is awkward if you use a wired keyboard and mouse.

Performance is well balanced by Intel's Core 3 304, though it has its limitations. It’s quick where it counts when browsing the internet, using office software, or playing media, and I've even found myself using it to offload image and video files to the server quickly and to book equipment and locations. Where it does fall is in graphics performance, as highlighted by the benchmarks: great for the office but really a no-go for gaming unless optimised for lower-specification systems.

Storage is also interesting with two M.2 slots. However, they support different lane counts, and a Crucial T710 Gen 5 drive returned 3510.25 MB/s read in slot one but only 1750.42 MB/s in slot two, which is slower than the UFS drive Beelink already fits, but then does enable you to boost the overall internal storage neatly.

Everything else is again well-balanced for office use. This is a nice, quiet machine with Windows 11 Pro pre-installed. The design is nothing flashy, but for dedicated office use with serious network potential it’s one of the best mini PCs I've tested.

Beelink EQi Wildcat Lake: Price and availability

• How much does it cost? From $509
• When is it out? Available now
• Where can you get it? Directly from Beelink and retailers such as Amazon

The EQi costs $509 for the 16GB LPDDR5 and 512GB configuration that I’ve looked at in this review. It's available direct from Beelink here in three configurations - 16GB LPDDR5, 24GB DDR5, and 32GB DDR5 RAM, all with 512GB SSD.

I'm also seeing several models on Amazon.com and Amazon.co.uk for around the same price.

• Value: 5 / 5

(Image credit: Alastair Jennings)Beelink EQi Wildcat Lake: Specs

CPU: Intel Core 3 304, 5 cores/5 threads, up to 4.3GHz, 6MB cache, 15W TDP
Graphics: Intel Xe3-LPG integrated graphics
RAM: 16GB LPDDR5, or 24GB/32GB DDR5 at 5600MT/s; DDR5 version upgradeable to 64GB
Storage: 512GB UFS 3.1; two M.2 2280 slots (slot 1 PCIe 4.0 x2, slot 2 PCIe 4.0 x1), up to 4TB each
Left Ports: USB 3.2 Type-A, USB 3.2 Type-C, 3.5mm audio jack
Right Ports: 2x Thunderbolt 4, 2x USB 2.0, HDMI 2.0, 10GbE, 2.5GbE, AC input
Connectivity: Wi-Fi 6, Bluetooth 5.2, 10GbE Ethernet, 2.5GbE Ethernet
Audio: 3.5mm combined audio connection
Camera: None
Size: 126 x 126 x 44.2mm,
Weight: 500g
OS Installed: Windows 11 Pro
Accessories: HDMI cable, regional AC lead/power adapter and user manual

Beelink EQi Wildcat Lake: Design

If you’re looking for a machine that blends into the office background, then the Beelink EQi Wildcat Lake is a perfect option. The dark grey finish, rounded corners, and a lightly textured top panel are neat and strike a note that really highlights the machine's price point and specifications.

The size is also in line with what you’d expect from a standard Office Mini PC, with a relatively small form factor and a compact 126mm x 126mm footprint, at just 500g.

While the design might not be the most inspiring, it is generally well thought out and, surprisingly, has a built-in 85W power supply, so just a single power cable runs to the AC mains, with no need for a power brick.

On the front and back are a good range of ports, highly specified for a machine at this price. There’s also good air ventilation through the base, with airflow entering over the processor and passing through the storage.

(Image credit: Alastair Jennings)

Beelink quotes 32dB as the noise in normal use, and this seemed about right, with the ambient noise in the office slightly higher when ideal. This is also one of the neatest laid-out Mini PCs internally that I’ve seen, accessible through the base, which is held in place by four small cross-head screws. A small attention to detail is the small label that just helps to lift the base free, enabling easy access.

Once inside the interior, both M.2 slots are slightly hidden under a single large metal heatsink that lifts off with two screw. Once removed, your choice of M. 2 2280 SSD can be installed. This heatsink shows real consideration as a thermal solution for the installed drives, with the installation process taking no more than a couple of minutes.

The port selection and specification are where things really get interesting. Starting at the back, there are two USB 2.0 Type-A ports on either side of the HDMI Max 4.0 60Hz output rather than together, so a wired keyboard and mouse end up separated by a display cable.

Then there are two LAN ports, a 2.5GbE and a 10GbE, which is very unusual for a machine at this level; also unusual are the two Thunderbolt 4 ports. The front is a little more standard, with a USB 3.2 Type-A, a USB-C 10 Gbps, a 3.5mm audio jack, and finally the power button.

• Design: 4.5 / 5

(Image credit: Alastair Jennings)Beelink EQi Wildcat Lake: Features

A quick check through the features highlights the intended audience for this PC. At the heart of the machine is the very respectable Intel Core 3 304; it’s nothing flashy, but this is a new-generation CPU so offers far more than entry-level CPUs in the past, and while not an AI powerhouse, does feature an integrated NPU as well as GPU, handy if you use Copilot on a day-to-day basis for work.

The CPU offers a five-core, five-thread processor with one performance core and four efficiency cores, 6MB of cache and a 15W TDP, turboing to 4.3GHz. Beelink quotes AI performance at 24 TOPS across the CPU, GPU and NPU for local AI processing.

While it’s great to see this entry-level office machine has AI potential, connectivity is what has surprised me about the EQi through this test. On the back are those two Thunderbolt 4 ports that run at 40 Gbps with DisplayPort 1.4, and if you need to connect more monitors, then there’s the more traditional HDMI 2.0, which I used to connect to my monitor for this review.

The two network ports, one 10GbE and one 2.5GbE, are a headline feature and enabled me to connect directly into the 10GbE LAN in the office. This meant that I could quickly offload video and image files through the machine's Thunderbolt 4 to the network. While this is an office-focused machine, I use an old Intel NUC 9 in the office for this exact task, and its speed over 1GbE and USB3.2 Gen 1 is unmatched by this.

Inside there’s a fixed 16GB LPDDR5 or upgradeable DDR5 at 24GB or 32GB, both at 5600 MT/s, with the DDR5 version supporting up to 64GB. Storage is 512GB of UFS 3.1, again fixed, but there is the option to install two M.2 2280 SSDs, both enabling up to 4TB. One point about these two slots is that they don’t support the same number of PCIe lanes, with one half as fast as the other.

Wireless runs over a MediaTek MT7922 module, with Wi-Fi 6 and Bluetooth 5.2.

(Image credit: Alastair Jennings)

• Features: 4.5 / 5

Beelink EQi Wildcat Lake: PerformanceBenchmark scores

CrystalDiskMark Read: 1977.4
CrystalDiskMark Write: 806.79
Geekbench CPU Multi: 6303
Geekbench CPU Single: 2337
Geekbench GPU: 6182
PCMark Overall: 4733
Cinebench CPU Multi: 4785 (R23)
Cinebench CPU Single: 1696 (R23)
Fire Strike Overall: 1918
Fire Strike Graphics: 1971
Fire Strike Physics: 12379
Fire Strike Combined: 778
Time Spy Overall: 879
Time Spy Graphics: 773
Time Spy CPU: 3963
Wildlife Overall: 5133
Steel Nomad Overall: 127
Windows Experience Overall: 8.1
Procyon AI Computer Vision 2.0: 635

Additional tests

Crucial T710 Gen 5 NVMe, slot 1: Read 3510.25 MB/s, Write 3530.98 MB/s
Crucial T710 Gen 5 NVMe, slot 2: Read 1750.42 MB/s, Write 1766.17 MB/s
Lexar PCIe 3.0 SSD: Read 1631.56 MB/s, Write 1499.14 MB/s

Getting started with the Beelink EQi was extremely straightforward, with a single power cord that plugs in, eliminating the need for the usual power brick. While this may seem small, it actually makes quite a difference, reducing cables and bulk to deal with.

Once the initial setup with the mouse, keyboard, and 4K monitor was complete and Windows 11 Pro was set up, the machine was installed with the usual array of applications. Microsoft Office, Adobe Creative Suite and a few games for good measure.

Starting with the main office applications, performance proved solid, running PowerPoint, Word, and Excel without issue, with Copilot there to support me as I put together a presentation for September. The speed is good, and despite the CPU's entry-level position, it handled everything impressively.

Checking out the results of the Geekbench 2,337 single-core and 6,303 multi-core results and, more importantly, the PCMark 10 came in at 4,733, showing that while this machine doesn’t excel, it’s where you would want it for a competent everyday computer that won’t get bogged down under heavy processing.

During the test, I edited a few images in Lightroom and Photoshop, and both ran surprisingly well, even with the images stored on the internal UFS 3.1 drive. While this drive isn’t the fastest compared with the latest SSDs, it is fast enough for most office-based work, with CrystalDiskMark results of 1977.4 MB/s read and 806.79 MB/s write.

What’s interesting is that inside there are two PCIe 4.0 slots, one listed as PCIe 4.0 x2 and the other as PCIe 4.0 x1, so I fitted a Micron Crucial T710 Gen 5 NVMe to each in turn to check out the speeds of each, just as a note you can add up to 4TB in each, giving a total of 8.5TB of storage which is impressive.

The first slot recorded 3510.25 MB/s read and 3530.98 MB/s write. The same drive in slot two returned 1750.42 MB/s read and 1766.17 MB/s write. That's roughly twice the speed in the first compared with the second and still very acceptable for the intended end user. While I used a PCIe 5.0 SSD, a respectable PCIe 4.0 SSD which is far cheaper should gain a similar result.

Switching to the Xe3-LPG graphics, things start to fall slightly. The integrated graphics are ideal for everyday work, watching back multimedia and light photo and video editing.

However, when it comes to gaming, the potential is limited. Hogwarts Legacy ran slowly, and Indiana Jones and the Great Circle was a no-go. Other games ran, but due to the lack of space on the internal storage, I ran all from the faster Crucial in the first of the two spare SSD slots.

After running the Mini PC in the office for a week, using it mainly for Word and research work with some light photo editing, a few features became apparent that set this Mini PC and test apart from most others. The speed of those connections just enabled the EQi to integrate perfectly with the studio; it’s nothing flashy, but it turned out to be the fastest machine for transferring image and video data to the NAS.

After a shoot, I plugged the CFExpress Type-B card reader into the Thunderbolt 4 port and transferred the files at speed over to the Ugreen IDX6011 NAS. The connection speed over the network, due to that 10GbE port, made the transfer speed far exceed that of any other machine I know.

I usually use an old Intel Nuc 9 Ghost Canyon that I can leave for a few hours before network editing on the MacBook Pro; however, here, with this small, inexpensive machine, it handled everything at speed.

While the 24-core TOPS, as quoted by Beelink, is marginal on the AI processing front, and while I was able to use Co-Pilot, running a local LLM was limited to a 3-4B model, I opted for Qwen3. The main issue here was that the RAM is only single-channel, which limits the speed; however, for checking through documents and basic summary tasks, it actually worked surprisingly well.

• Performance: 4 / 5

(Image credit: Alastair Jennings)Beelink EQi Wildcat Lake: Final verdict

(Image credit: Alastair Jennings)

The Beelink EQi is an impressive office-focused machine that underwhelms with its design but outperforms in performance. It packs in the usual office-based Mini PC features such as being able to support three displays, 2.5GbE networking, a base level of storage and RAM, but then it surprises by offering more.

The 10GbE networking and Thunderbolt 4 ports are surprising and completely change what’s possible with this machine, and for something with such a small footprint, those two specifications instantly make it interesting.

When you then add in the fact that you can expand the internal storage with two 4TB SSDs, it then becomes even more interesting.

For me, the 10GbE and Thunderbolt ports are what make this Mini PC go from something that would blend into the background to something that’s actually very useful. The speed of those two ports enables me to hook this directly into an existing 10GbE network and to take advantage of the speeds that it offers. The issue, of course, is that you need the network infrastructure in place to make the most of this speed.

That brings me round to who the Beelink EQi is best for? Straight down the line, it would be for Office work. It has excellent single-core performance, runs pretty much silently, supports up to three displays, lets you upgrade the internal storage, and has an ultra-fast network connection that is beyond what most desktops offer. It’s also not really usable for playing games, and, on the surface, if you work in the creative sector, it’s not going to make an impact there either.

However, as I found with a cheapish mini PC that offers solid performance, the Thunderbolt 4 and 10GbE network make it great for offloading large video and image files to the NAS. Realistically, however, this is a Mini PC perfectly suited to anyone whose work involves documents, spreadsheets, browsers, and moving large files across a network; it is one of the more capable small machines I have tested this year.

Should I buy the Beelink EQi Wildcat Lake?

Value

Decent connection options for the money, if you have a 10GbE network to use it

4

Design

Compact, quiet and well laid out inside, let down by plasticy casing

4

Features

Outstanding port selection for a machine this size, focused for the office but with potential

5

Performance

Fast in everyday work, weak in graphics, excellent storage in the right slot

4

Overall

An excellent quiet office machine with workstation connectivity

4

Buy it if...

You store a lot of files in different places

Whether that is a NAS, a media library, or a shared project drive, the dual 10GbE and 2.5GbE ports move large files at a rate most full-sized desktops can’t match.

You want a silent machine.

Single-core performance is excellent, and in the office I couldn’t hear the machine over the ambient noise in most situations.

Don't buy it if...

You want to play modern games.

A Time Spy graphics score of 773, compared to 3,963 for the CPU, shows exactly where the limitations are. The single Xe3 core is built for displays, video playback and light graphical work, not gaming.

You need heavy multicore performance.

While light video and image editing is possible, it’s certainly not where the machine's strength lies.

For more professional-grade desktops, we've tested the best business computers.

• A spent Falcon 9 upper-stage rocket will hit the Moon near the Einstein Crater on 5 August at 5,400 mph, releasing roughly the energy equivalent of three tons of TNT
• The impact, though unintentional, was predicted since September 2025, giving NASA's LRO and South Korea's Danuri time to plan before-and-after imaging of the new crater
• Its scientific value is a rehearsal rather than a measurement: no seismometers have operated on the moon since 1977, so the event mainly tests impact-localization methods for future experiments

At roughly 06:35 UTC on August 5 2026, a spent Falcon 9 upper stage will stop being space junk and start being a crater.

The object, cataloged as 2025-010D, is projected to strike the lunar surface near the crater Einstein at 2.43 kilometers per second, which converts to about 8,700 kph or 5,400 mph.

While this was not intentionally aimed at the moon, almost a year of advance notice separates it from the usual category of orbital accident, even as NASA's Lunar Reconnaissance Orbiter and South Korea's Danuri are being prepared to take before-and-after images of the area on the moon where it is expected to make its impact.

An increasingly crowded atmosphere

The object itself is the spent upper stage of the Falcon 9 which launched from Kennedy Space Center on January 15, 2025, carrying two commercial landers: Firefly Aerospace's Blue Ghost Mission 1, which went on to make the first fully successful private lunar touchdown that March, and ispace's Hakuto-R Mission 2, which was lost shortly before landing in June.

Independent astronomer Bill Gray, whose Project Pluto software produced the original prediction, puts the stage at roughly 12 meters long, about the height of a five-story building, with a mass near 4,900 kg.

After separation, it stayed in an elongated Earth orbit with a 26-day period, swinging between roughly 220,000 and 510,000 km. Its path and the moon's path cross, but for 18 months the two objects have never met at the crossing point.

Gray expects a single crater around 17 meters across, reasoning by analogy with the double crater left by a Chang'e-5 T1 upper stage in 2022. The observation-planning paper led by Benjamin Fernando points higher, at roughly 27 meters across and 5 meters deep. Either way, the impact should release energy equivalent to about three tons of TNT.

There are limits to what can be measured from it. When NASA deliberately crashed Saturn V third stages and lunar modules into the moon between 1969 and 1972, those impacts were seismic sources fired into an emplaced network of four seismometers. That network was switched off in 1977, and nothing has replaced it yet. So while orbiters will photograph the resulting crater, the impact cannot be recorded seismically at all.

This will be only the second dead rocket known to have hit the moon by accident, after the 2022 Chang'e-5 T1 event, which China has never publicly acknowledged. Gray, for his part, says the episode points to a certain carelessness about how leftover space hardware gets disposed of. Objects in low Earth orbit eventually meet the atmosphere and burn up. The moon has no such cleaning mechanism, so lunar impacts are permanent.

• Nothing's cheaper CMF arm just launched CMF Clip Pro, its first clip-on buds
• LDAC, new '3-point Clip Design', 10.8mm driver and just 5.9g per bud
• £79 / $99 / AU$179 with no flagship 'Audio Snapshot' or call recording — and I'm glad

Loved Nothing until recently, but now you’re not so sure? If you looked at my recent Nothing Ear (3a) review and thought "Wait, they record calls? Why?" you're not alone.

Thankfully, the firm’s cheaper CMF arm just launched its first-ever clip-style open earbuds with LDAC and the basics — ie., no in-call recording — and they're delightfully chic and affordable. I predict a return to form.

CMF says the Clip Pro (for that is their name) are designed to deliver a listening experience that "feels as natural as it sounds" and that the buds are built for "people who move through busy cities, long workdays and active lifestyles".

Open earbuds of this type are growing rapidly in number, but achieving comfort, the ability to hear your surroundings (to be clear: the 10.8mm driver in each earpiece rests in the well of your outer ear, like an ear cuff), and decent sound quality is one of this category's biggest challenges. With Clip Pro, CMF says it set out to solve these issues.

How? First, it starts with a new 3-point Clip Design, which the company describes as a "carefully balanced three-point support system that distributes pressure naturally across the ear instead of concentrating it in a single area". Lovely stuff.

CMF Clip Pro: what you need to know

(Image credit: CMF)

The design features an "elastic titanium wire core" with a "slim C-bridge" to fit your shell-like, and at 5.9g per earpiece, a barely-there wearing experience is the goal.

In terms of audio chops, the 10.8 mm dual-magnet dynamic driver in each earbud is aided by CMF's own Ultra Bass Technology, which the company says "intelligently enhances low frequencies while simultaneously reducing treble distortion" for a more engaging experience through those tricky-to-nail-in-this-design lower frequencies.

Clip Pro supports LDAC, for higher-quality wireless streaming. It's not the first time we've seen the better-quality codec supported in Nothing's affordable earbuds, but it's welcome at this entry-level price if you own a Samsung or Sony phone.

Think they'll be a bit hit-and-miss for calls? CMF thought of that too. Clip Pro combines what the company calls its "flagship VPU-powered Clear Voice Technology and a four HD mic system" to separate the wearer's voice from surrounding noise in real time.

For users who value privacy (and I'd say that's all of us), CMF says its Sound Seal system should reduce sound leakage by intelligently managing mid and high frequencies. The goal? Making conversations and music less audible to people nearby.

All of this can be customized in the Nothing X app (a companion app I like very much, having tested most of Nothing's output to date), with dedicated settings for calls or both calls and audio playback.

Oh, and Clip Pro also totes CMF's signature smart dial, as seen in the CMF Buds Pro 2. It's a tactile rotating control built into the charging case for adjusting volume, managing playback, answering calls and pairing devices without needing to unlock your phone. We were wary of the implementation previously, summarising in our review of those July 2024-issue buds: "The dial is prone to accidental turns or presses in your pocket, the notification sound to tell you that a function has been triggered is startlingly loud, and there’s a significant delay between using the dial and hearing its result", so it'll be interesting to see if CMF has tweaked its efficacy here.

You also get on-ear physical controls (which the company says will work even if the user is wearing gloves) as well as a low-latency mode for gaming, dual device connectivity and Nothing's 'find my' solution, which makes your lost earbud emit a rattlesnake-like sound to help you find it under the couch, say.

Your color options here are Dark Grey, Light Grey, and Coral. In case I haven't stressed this enough, you should not wear all three hues like the model in the main image of this missive. The buds are also IP54 rated for water and dust ingress.

Stamina? Of course: Clip Pro's battery claim is up to 10 hours of continuous playback on a single charge and up to 32.5 hours of total playback with the charging case.

Let's talk money: £79 / $99 / AU$179, and they'll hit shelves on August 15, 2026. We can't vouch for them yet, but we'll be working on it…

The Telegram CEO says Apple made the app unavailable for download without any warning.

• Whoop has partnered with women's health tracking app Natural Cycles
• Whoop users get one year of the app totally free
• Natural Cycles will use that data to provide a 'daily fertility status' through its app

Whoop has partnered with women's health tracking app Natural Cycles, offering a year's worth of membership to existing Whoop users free of charge.

Whoop users also using the Natural Cycles app can now enable their Whoop app to import skin temperature data into Natural Cycles, said to 'eliminate the need for manual daily tracking' using a fingertip sensor or similar. Natural Cycles will incorporate this data to provide a daily fertility status, as basal body temperature is a key indicator in cycle tracking.

“At Natural Cycles, we're committed to making digital contraception as seamless as possible for our users," Dr. Magda Armbruster, VP of Product at Natural Cycles, said in a statement.

"Integrating with WHOOP removes friction from the experience by allowing members to use the wearable they already rely on to automatically capture overnight skin temperature used by the NC° app to determine daily fertility status."

It's worth noting the integration is only available with Whoop MG and Whoop 5.0 devices: anyone still rocking the Whoop 4.0 is unfortunately left out in the cold.

(Image credit: Whoop)

Whoop isn't the only wearable that Natural Cycles works with. Natural Cycles can use data from Whoop's fellow subscription-powered premium wearable, the Oura Ring, as well as the best Garmin devices and Natural Cycles' own NC Band.

Natural Cycles costs $149.99 in the US, £89.99 in the UK or AU$149.99 in Australia, so you're saving a packet here.

Natural Cycles is unusual for a women's health tracking app in that it claims to be the 'only FDA-cleared, hormone-free birth control app'. Clearance from the US Food & Drug Administration is a badge of quality in the consumer health tech space.

The Whoop's recently seeing a new challenger in the screenless smart band form factor: the Garmin Cirqa. You can watch our video review of the device below.

The feature — previously rumored as Pixel Glow — may bring a visual twist to Google’s next Pixel series

These shows and tons of great new movies arrive to HBO Max this month.

• Cheap smart glasses are selling out as people look for Meta Ray-Ban alternatives
• More camera glasses are leading to more privacy concerns for many
• I wouldn't buy them for plenty of other reasons

If you want a pair of smart glasses, you’re spoiled for choice in terms of what they can do technologically, and over time we’ve seen more and more budget options appear. These cheaper models are the first step in making emerging gadgets more accessible, but for smart specs there’s a fear they might simply exacerbate the growing concerns over the tech.

Australian discount store Kmart recently went viral after the AU$89 Anko glasses it was selling sold out entirely (via The Guardian). In the US and UK, you don’t have to look far to find equally affordable camera glasses available through reputable digital marketplaces such as Amazon.

Given the proven popularity of Meta’s Ray-Ban and Oakley specs, it’s no surprise the market would see an influx of smart glasses, especially cheap pairs that aim to mimic much more expensive rivals while keeping costs low — typically by downgrading their technological capabilities and not partnering with premium design brands.

Premium smart specs, at a premium price (Image credit: Ray-Ban / Meta / Coperni)

However, cheap smart glasses aren’t as universally celebrated as cheap TVs or affordable smartphones — scrolling through online forums discussing this news, the replies aren’t kind about smart glasses, discussing the pitfalls of smart glasses in general and more affordable options specifically.

The most prominent concern is that cheap camera glasses expand the range of people with the technology. More people with these glasses means more people possibly recording people in public without their knowledge.

There are also fears that cheap camera glasses might not come with the same sophisticated tamper protection glasses from more recognizable brands boast. Either through tape coverings or minor electrical modifications, folks could disable the camera light and use the specs even more covertly than they otherwise would — they’ve tried the same with Meta specs, but Meta has patched out these workarounds so far.

Privacy and other problems

(Image credit: Getty Images)

Beyond the impact on the general public, cheap smart glasses users might want to consider the privacy implications of wearing cameras from unknown brands.

We’ve seen that even super well-known brands like Meta aren’t perfect — for example, there are reports that contractors could see videos (including intimate or otherwise private content) taken using the glasses if it involved the AI in some way — so why should we expect completely unknown brands with nothing to lose to be any better with privacy?

In fact, considering their lack of notoriety makes them less likely to get investigated in the way Meta was by journalists who uncovered the sharing of videos with contractors. The ease with which tiny brands can dissolve and pop back up under different names, I for one would be a lot less trusting of what they do with my data and of their phone apps.

Of course, many lesser-known brands behave responsibly, but when there’s so much room for things to go wrong, for private information or video to get leaked, I just have this extra layer of paranoia I can’t shake.

Want smart glasses? Go with someone trusted (Image credit: Future)

Even ignoring my personal privacy concerns, as someone who has tested countless smart glasses at this point, I’ve found that better-known brands are simply better at making hardware and software.

I’ve worn so many uncomfortable specs with tinny audio, awful camera quality, terrible AI, and apps that are worse than useless. They’re sluggish. They’re ugly. They’re not worth even a budget price.

Just don’t buy them.

If you do want smart specs, save up for something proper — it doesn’t have to be from Meta, but I’d recommend choosing a brand with plenty of solid independent reviews. While I’m using them less for reasons better explained by other writing, I have gone weeks on end wearing a good pair of smart glasses thanks to comfy yet fashionable design and solid tech. Whereas the cheap (and terrible) glasses I’ve tried I’ve wanted to rip off my face within hours.

You get what you pay for.

The company now operates its autonomous ride-hailing service in more than a dozen cities.

You can play the Gears of War: E-Day beta, Ball x Pit and more this month.

• Amazon's Claude Sonnet project cost $1.8 million, 860% over budget
• The overspending remained undetected internally for nearly five months straight
• A financial auditing tool project exceeded its budget by $541,000

Amazon has confirmed an internal Claude Sonnet deployment intended for matching author details with product listings ballooned far beyond its planned budget.

The Financial Times found the project ultimately cost the company $1.8 million, marking an increase of 860% over the original allocation.

The overspending went undetected for roughly five months, raising fresh questions about how closely Amazon tracks its growing AI expenditures internally.

Mounting costs across multiple projects

Amazon's Claude-related overspend was not an isolated incident within the company's broader push toward deploying autonomous coding agents across various internal teams.

A separate project meant to build a financial auditing tool reportedly exceeded its budget by $541,000, compounding concerns about unchecked automated spending.

Another logistics-focused initiative, designed to shorten delivery times across Amazon's broader distribution network, incurred an additional $134,000 in unplanned costs.

Company insiders explained that mistakes once considered trivially cheap have grown catastrophically expensive as token-based pricing models replaced older subscription arrangements.

This shift coincides with AI agents gaining broader autonomy, allowing errors to multiply quickly before human reviewers noticed the growing financial damage.

Amazon pushes back on the narrative

In response, Amazon issued an internal statement acknowledging ongoing experimentation while disputing characterizations of these incidents as routine business practice.

"As with any new technology, we're experimenting, learning and improving how we use it, including how we drive cost efficiencies," the company said.

The company further stated that isolated examples portrayed as business as usual do not accurately capture how its teams actually apply AI tools.

Amazon's quarterly revenue exceeds $181 billion, meaning the disclosed AI overspending accounts for less than 0.1% of one month's earnings.

This pattern of AI-driven cost overruns follows earlier troubles at AWS, where automated coding bots triggered several unexpected service outages this year.

Amazon responded to those outages by restricting AI agent permissions rather than granting them access equal to senior engineers overseeing similar tasks.

The company also discontinued an internal leaderboard once used to track employee AI usage, as rising costs prompted a reconsideration of that approach.

Executives across the industry continue framing AI spending as a necessary investment despite growing evidence of inconsistent returns.

Amazon's continued experimentation with AI-driven coding tools comes despite skepticism from other corners of the tech industry regarding its actual business value.

Uber's chief technology officer, whose company also operates a substantial logistics network, has publicly stated no clear connection exists between heavy AI adoption and successful software delivery.

This skepticism stands in contrast to Amazon's own internal messaging, which frames the company's approach as ongoing experimentation rather than a fully validated cost-saving strategy.

Some analysts now argue that unchecked automation could quietly erode profit margins long before leadership notices meaningful financial impact.

For a company of Amazon's scale, these overruns remain financially minor, though they suggest oversight gaps that smaller companies may struggle to absorb.

If you’ve played at least two EA Sports-made football games in the past 10 years, you’d be forgiven for thinking the developer routinely wheels out the same basic product in a fresh lick of paint. New kits, up-to-date transfers, and minor gameplay tweaks have been enough to keep millions of players coming back to the same core FC experience, but in FC 27, EA Sports is introducing something truly, genuinely new (no, really this time).

The Grounds is essentially the developer’s take on NBA 2K’s The City — an open-world reimagining of its Clubs mode in which players can compete with friends or randoms in both stadium- and street-based challenges.

In EA Sports’ own words, the fantasy of The Grounds is as follows: “You create your footballer, explore a football world, play football, and progress your footballer to the very top. From casual drop-ins with other players in the parks to the heights of competitive Clubs play in the stadiums, The Grounds is your destination for avatar- and Clubs-based social football [...] It's a football playground built for you and your friends.”

Remember those halcyon days when you used to meet your mates for a kickabout after school? That’s the vibe EA Sports is attempting to capture here.

The Grounds themselves are three distinct districts, each representing a different “cultural hub of football.” There’s Parkside, whose grass pitches and terraced brick houses are reminiscent of small-town England; Zeiza, a colorful, Argentina-inspired suburban hub; and Montclair, a Parisian-style metropolitan area featuring 4G turf and hard surfaces.

These areas will be populated by other online players and NPC ‘mentors’ including Kylian Mbappé, Paulo Dybala, Chloe Kelly, and Alex Hunter (yes, that Alex Hunter), who will initially teach you the mechanics of The Grounds and later task you with specific challenges outside of competitive matches.

The Grounds map in EA Sports FC 27 (Image credit: EA Sports)

Speaking of competitive matches, you can choose to engage in traditional, stadium-based 11v11 matches in The Grounds, or jump into more casual, small-scale modes like 1v1, 2v2, 3v3, and 4v4. There are also new party-style Kickabout modes to try, including dribbling-based obstacle courses and the basketball-inspired Bucket Ball.

And if you’d rather not play against anyone at all, then Drop Kick Targets are your friend. These randomly spawning neon circles are essentially floating goals to be targeted while you walk down the street — think Odin’s Ravens in God of War. Or Riddler Trophies, but less rage-inducing.

Completing matches, Kickabouts, or Drop Kick Targets earns you Coins and Archetype XP, which can be used — surprise! — to level up your virtual pro in The Grounds. As well as new Playstyles and stat boots, you’ll also be able to buy fully licensed virtual clothing from brands like Nike, Puma, and New Balance, which sounds a little icky (the Metaverse died for a reason, guys!), but virtual currency has been a feature of every EA Sports game for the last two decades, so its inclusion again here was a given.

Kylian Mbappé is one of four mentors featured in The Grounds (Image credit: EA Sports FC 27)

I’ve never been much of a Clubs player, but I’m glad EA Sports is finally giving this long-neglected mode the attention it deserves in FC 27. Sure, The Grounds looks like a blatant copy of NBA 2K’s The City mode, but it’s undoubtedly a breath of fresh air for the FC franchise specifically, which has thus far struggled to establish its own post-FIFA identity.

I’m less convinced about The Grounds’ staying power alongside tried-and-tested modes like Ultimate Team and Career. Its three districts and minigames look like they’ll be fun to explore for a time, but will players really keep returning to the streets to complete Mbappé-issued challenges, kick balls into a basketball hoop, or visit a virtual clothing store long after the game’s initial release?

That, of course, will be the developer’s problem to solve — but hey, if there's one thing EA Sports knows how to do, it's keep players coming back.

Fanxiang PS2000W: 30-second review

Fanxiang is a Chinese storage brand, most commonly known as Shenzhen Fanxiang Information Technology. The company grew out of a team with a decade of prior experience in storage design, firmware, and product sales, and they've built their range around four main product lines: solid-state drives, mobile storage, memory modules, and storage cards.

It’s been making external SSD storage for about five years, with a wide range of devices and specifications.

The PS2000W appeared in 2025, and delivers capacities from 500Mb to 4TB in a slim aluminium case. It comes with a soft carry pouch and a cable that has an integrated adapter for both USB-A and USB-C connections.

The quoted performance is 2000MB/s reads, and 1800MB/s writes, but that assumes you have a USB 3.2 Gen 2x2 port, and that’s not a common standard.

More realistically, a USB 3.2 Gen 2 port will achieve a little over 1000 MB/s reads, and 950 MB/s writes with either a USB-C or USB-A connection.

For those wanting faster transfers, you either need a system with the rare USB 3.2 Gen 2x2 port, or choose a different drive that supports the USB4 standard, like the Fanxiang PS3000 portable SSD.

However, that option costs $350 for the 1TB model, which is substantially more than the $150 the PS2000W is typically priced at for that capacity.

The problem the PS2000W doesn’t address is that this drive is double the price it was only eight months ago, and unless you have a system that can make the most of its enhanced performance, there are cheaper alternatives. In fact, if you have a spare NVMe drive, then buying an external enclosure for USB4 is probably a better choice.

Not the best portable SSD, or the worst, but sadly impacted by the outrageous cost of flash memory in 2026.

(Image credit: Mark Pickavance)Fanxiang PS2000W: Pricing and availability

• How much does it cost? Starts from $90
• When is it out? Available now
• Where can you get it? Sold through online retailers and also directly from Fanxiang.

The Fanxiang PS2000W portable SSD comes in capacities starting at 500 GB, doubling each step to include 1TB, 2TB, and 4TB drives.

Directly from the maker, these are priced at $112.99, $203.99, $379.99 and $495.99 for the largest capacity. That pricing makes the 4TB the best value, and the 500MB the worst.

But the maker prices aren’t especially good, and I was able to find it much cheaper on Amazon. On Amazon.com in the US, the 500MB model is only $89.99, the 1TB model is $149.99, and the 4TB model is $419.99. There wasn’t a price for the 2TB version.

UK pricing via Amazon.co.uk is £64.60, £119.99, £169.99, and £319.99, and Amazon EU sites are €94.99, €139.99, €229.99, and €379.99, respectively.

For those curious, the UK is the cheapest place to buy this hardware, followed by the EU, and the most expensive is the USA.

While officially Crucial has exited the retail market, its equivalent product, the X10, is still available from many retailers. At $207 for the 1TB option, and $299.99 for the 2TB option, the X10 is slightly more expensive. But it is rated as water-resistant, and while there isn’t a 500GB model, there are 6TB and 8TB options.

A cheaper option with the same spec is the PNY RP60, but the 1TB model is still $10 more than the PS2000W.

But many of the familiar brands are more expensive, such as the SanDisk Extreme Pro V2, Samsung T9, Kingston XS2000 and LaCie Rugged Mini. Western Digital has its My Passport SSD, but it’s the same price as the PS2000W, but only offers half the performance.

Those looking for a cheaper option should consider the $82 Orico S20 with 512GB storage.

It’s tempting to look at the pricing of the PS2000W and see it as a good deal, when it’s far from that.

Price tracking the 1TB model: it was $149.99 at the time of writing; it was $75.99 in November of 2025. That’s nearly a 100% markup.

Now, I should say that Fanxiang doesn’t control the price of Flash memory, which has spiked since AI companies and speculators decided to buy all of it. But in good conscience, I can’t recommend people spend this sort of money on devices that cost half as much a year ago.

• Value: 3 / 5

(Image credit: Mark Pickavance)Fanxiang PS2000W: Specs

Model

Fanxiang PS2000W

Interface

USB 3.2 Gen 2x2 (20Gbps)

Sequential read

Up to 2050MB/s

Sequential write

Up to 1800MB/s

Flash type

TLC 3D NAND

Capacities

500GB, 1TB, 2TB

Dimensions

8.5cm x 5cm x 0.8cm

Weight

Approximately 45g

Build

Aluminium alloy unibody

Cables included

USB Type C to C, plus Type C to A adapter

Compatibility

Windows, Mac, Android, PS5/PS4, iPhone 15 and later via USB-C

Warranty

5 years

Fanxiang PS2000W: Design

• Aluminium enclosure
• Dual USB standards cable

Some brands, and you know the names, will try to make their external SSDs and HDDs as cheaply as possible, typically by using mostly plastic construction.

Thankfully, Fanxiang didn’t go down that path with the PS2000W, and this device appears to be entirely made of metal. And, based on its relatively low weight, it uses aluminium.

The package is a simple lozenge shape with curved sides and flattened ends, measuring 85mm long by 50mm wide and less than a centimetre thick.

Out of the box, it comes with a 30cm USB-C to USB-C cable that has a USB-A adapter permanently attached. The drive and cable fit neatly into a soft drawstring bag, also provided, making for a neat and easily pocketable solution.

The review model was a 2TB capacity drive, which translated into 1863GiB of usable space, and it connects to the host device using USAP (Serial ATA).

To achieve the best performance, it requires a USB 3.2 Gen 2x2 port, but the drive is backwards compatible with USB 3.2 Gen 2, Gen 1, and even USB 2.0. However, as you move down the USB interface pecking order, the expected speed decreases.

As for features, this drive is basic. It supports S.M.A.R.T. NCQ, TRIM and GPL, but it doesn’t offer hardware encryption like on the Kingston XS2000 or the Crucial X10 Pro.

(Image credit: Mark Pickavance)

There are also no software tools included, though you can download a utility for SSD management from the Fanxiang website. I didn’t notice any means to update the installed firmware, or any firmware releases.

To be blunt, this is a basic USB SSD with a Gen 2x2 interface, but otherwise it’s unremarkable. And therefore, if you don’t have that interface, there appears to be no compelling reason why you would choose this one over another brand.

(Image credit: Mark Pickavance)

• Design: 3.5 / 5

Fanxiang PS2000W: Performance

• Needs USB 3.2 Gen 2x2
• Performs within quoted specification

Fanxiang PS2000W Benchmark

Bench

Test

USB 3.2 Gen 2x2

USB 3.2 Gen 2

CrystalDiskMark 9.03

Default Read

2002MB/s

1041MB/s

Default Write

1853MB/s

948MB/S

Real World Read

1336MB/s

993MB/s

Real World Write

1219MB/s

912MB/s

AJA System Test 64GB

Read

1804MB/s

971MB/s

Write

1645MB/s

891MB/s

AS SSD

Read

1771MB/s

971MB/s

Write

1496MB/s

879MB/s

ATTO

Read

1.89GB/s

994MB/s

Write

1.73GB/s

906MB/s

PCMark Data Drive Bench

Score

2028

1250

Let’s be brutally honest: USB 3.2 Gen 2x2 was a dead standard walking from its inception, and USB4 didn’t really include it in its repertoire of supported prior standards. There are a few exceptions, but in general, if you don’t actually have a port to the 2x2 standard, then you won’t see the true performance that this drive can muster.

For that reason, I’ve presented the performance via a Gen 2x2 port and a much more common Gen 2 variety. How does it compare with the best Gen 2x2 drives?

Well, read speed is below the Crucial X10 Pro, ADATA Elite SE880 and LaCie Rugged Mini SSD, but write speed is close to what all those achieve.

Given the different controllers and implementations of Gen 2x2 interfaces, the Fanxiang PS2000W portable SSD doesn’t have a compelling reason to be chosen over any other drive in Gen 2x2 mode, but it can get close to the maximum realistic speed for this interface.

But if speed is the prerequisite, then a USB4 or Thunderbolt external drive, like the Corsair EX400U or the SanDisk Extreme PRO with USB4, would be a better choice if you have a USB4 port. These drives can hit read speeds of 3800MB/s or higher, and 3650MB/s writes.

And, if I were rushing to catch a plane or train, and needed to transfer files, I know which I’d prefer.

• Performance: 4 / 5

Fanxiang PS2000W: Final verdict

If the Fanxiang PS2000W portable SSD had appeared during the golden age of Gen 2x2, when the Kingston XS2000 appeared back in early 2022, and at a reasonable price, then it might have been something I’d recommend.

And the fine irony of that comparison is that the price of the PS2000W is almost exactly the same as the XS2000's MSRP was nearly four years ago. And the Kingston device supported hardware encryption, where the Fanxiang does not.

To be clear, the construction quality and performance of the PS2000W are absolutely fine, but what isn’t is the amount of money being asked for.

If money isn’t an issue for you, I’d buy the Crucial X10 Pro before they’re extinct, as it has hardware encryption. Or I’d consider a USB4-capable drive, like the Corsair EX400U, if you have a port that supports that specification.

Should I buy the Fanxiang PS2000W?

Value

100% more expensive than it was 8 months ago

3 / 5

Design

Metal construction, comes with a carry pouch and a dual standard cable

3.5 / 5

Performance

Works well at Gen 2x2 speeds, if you have that port

4 / 5

Overall

A nice drive, but hardly great value

3.5 / 5

(Image credit: Mark Pickavance)Buy it if...

You have USB 3.2 Gen 2x2
If you have a USB port that is Gen 2x2 (not just Gen 2), then this might be the drive you want to exploit the extra speed that it offers. However, unless all the ports you are likely to plug it into are this spec, the advantage will be limited.

Money isn’t a concern
Buying RAM, storage or GPUs is currently only for those people not paying with their own money. Since, depending on what you are buying, the price hikes range from 100% up to 400% and higher, if you can wait for the AI bubble to burst, you could save yourself, or your business, lots of money.

Don't buy it if...

You have USB 4.0
If you have USB4 or Thunderbolt 4, then this isn’t the drive for you, at whatever price. There are USB4-specified drives that are twice the speed of this one, and most of them only cost a little more.

You need bigger than 4TB
The largest drive available in this range is 4TB. And those wanting 8TB or more need to look for a different product entirely.

Also consider

Nextorage NX-PS1PRO Series
One of the first USB 4.0 capable external SSDs available, the Nextorage NX-PS1PRO Series comes in 1TB, 2TB or 4TB capacities.

Read our full Nextorage NX-PS1PRO Series review

OWC Envoy Pro FX
If you have Thunderbolt 3 ports or intend to add this functionality soon, then the Envoy Pro FX might well be worth your time. A beautifully engineered external storage option that can provide exceptional performance in the right circumstances, the OWC Envoy Pro FX is effective if expensive.

Read our full OWC Envoy Pro FX review

We've also tested out the fastest external SSDs around.

We have officially reached the finger-pointing phase of the blockbuster Apple vs. OpenAI lawsuit in which Apple alleges that, with the assistance of former Apple employees, OpenAI is stealing Apple trade secrets.

What the suit alleges, which I outlined here, is pretty damning: A pair of former Apple employees — one a 25-year company veteran and another who spent a decade as an iPhone electrical engineer — allegedly held onto systems, asked OpenAI interviewees from Apple to share proprietary Apple info, and, Apple claims, told partners that sharing Apple's proprietary manufacturing information was okay with the Cupertino tech giant.

OpenAI quickly denied the allegations and, in short order, we learned that Apple's claims that it had contacted Lui Chang about the charges were incorrect because Apple's counsel had mixed up the names of Lui Chang with OpenAI's own counsel, Che Chang.

This week, however, OpenAI offered a more robust response, releasing a text chain between Lui Chang and a current unidentified Apple employee, and the email exchange in which it's revealed that Apple had been talking to the wrong Chang.

"Apple is one of the greatest companies of all time, and built a reputation for obsessing over the smallest details. This careless, aggressive and oddly personal lawsuit sadly doesn’t live up to that reputation," wrote OpenAI on its company blog on Monday.

A cry for help

What the messages appear to reveal is not corporate espionage, but something far more mundane and, in some ways, very relatable if you've ever worked at a large company.

I should preface this to say that Apple has not confirmed these messages, but it's hard to imagine OpenAI fabricating them. In fact, the sheer amount of redacted information gives more credence to their authenticity.

There's so much missing information that it might seem impossible to glean anything from the banter between "Apple Employee #1" and Chang, so, to help myself, I went through and inserted guesses about '[Redacted - Apple Information]" to try and make sense of the conversation. (I decided much of it was about the potential use of Silicon Carbon batteries in the upcoming iPhone 18.)

In the messages, the Apple employees seem to have worked with Chang and were not only missing him, but his expertise.

"Just in case you don’t have enough work, I need some help. I have a vague recollection of you talking about a [REDACTED]. Can you give me a brief refresh of this and point me to an EE who may have some knowledge?" wrote the Apple Employee. I'm assuming he's referring here to an "Electrical Engineer"

Without apparent hesitation, Chang replied, "I think what you referred to is [REDACTED]. And the small [REDACTED]." He then pointed the Apple Employee to someone else at Apple who "might still remember anything."

How about, just say no?

Now, Chang could've said, "Hey, dude, miss you, too, but I can't really talk about anything Apple-related anymore." According to this exchange, Chang did not and kept doing his best to help.

This Apple employee also seems unaware of best business practices and keeps talking about technology, at one point even mentioning a schematic. It looks like he also shared that schematic with Chang, but it may also have been Chang sharing the image with the Apple employee.

As the Apple Employee presses on, Chang tells him to look for a schematic in the "team box folder". He even mentions a "power block diagram." I'm a bit surprised that bit wasn't redacted.

One telling section from Chang jumped out at me, and this is the relatable part. As Chang kept trying to connect Apple Employee 1 to the expertise he required, Chang noted, "Now when I think about it. The other smart people also just left that team. The one that might still [be] in Apple is [Apple Employee #10]."

Brain drain is a real issue, it seems, even for Apple. So in addition to long-timer Chang leaving for OpenAI, others who seem to possess some important technical information have departed as well.

Hi, this is highly irregular, please remove me from this thread."

Possibly an Apple employee

Later, as the quest for information continues, Chang recommended contacting a pair of Apple employees (#5 and #7).

At this point, there's a message that might've come from an automated system inside Apple's communication platform:

"Hi [Apple Employee #1 Name], [Apple Employee #3 Name], I can help get the conversation with [Apple Employee #6] and [Apple Employee #7] started."

Which led to my favorite part of the whole exchange. One of the Apple Employees auto-invited to the conversation essentially freaked out:

"Hi, this is highly irregular, please remove me from this thread."

Honestly, I feel for that person. As I've written before, Apple doesn't mess around when it comes to NDAs, proprietary information, or corporate secrets. Stumbling on a sort of illicit conversation about tech details between a current and former employee is surely enough to make any Apple employee hit the panic button.

Apple Employee #1 explains he added other employees "for awareness" but also shifts the conversation to an internal discussion. Chang is left to like the last message and, it seems, leaves it at that.

What's next?

The release of all this communication does not end the case. Far from it.

As of Monday evening, Apple filed with a US judge a strongly worded request for an injunction that, among other things, calls for "a preliminary injunction to stop the theft of its trade secrets.” It also includes the prediction, "Apple is likely to succeed on the merits of its trade secret claims."

Even so, I think this motion and the OpenAI post may move everything along toward an out-of-court settlement, one that would end the case but likely do little to repair this now-in-tatters Apple-OpenAI relationship.

In the end, though, OpenAI's post appears to reveal a mortal Apple, one that, despite its size and power, is still run by fallible humans who are often adrift, as any of us would be, when colleagues leave, taking with them friendship and, more importantly, knowledge.

Apple is building complex products, and corporate memory is a critical part of the process (so much so that incoming CEO John Ternus apparently just rehired a retired Apple hardware executive).

Understanding when, how, and why decisions were made and how they could impact future projects is invaluable. Sometimes, only those who were in the room know the truth of the matter. Chang, it appears, was one of those people. He had the information, took it with him, and now, it seems, someone wanted to keep the lines of communication open, which is exactly what Apple doesn't want.

Newly released texts and emails give us insight into how lawyers for two billion-dollar tech companies can’t figure out how to use their email.

• Bitdefender reported Roblox players lured by a fake “undetected” Xeno Executor mod spreading malware
• Infection chain delivers a Java‑based RAT and infostealer stealing browser data, online accounts, payment info, and crypto wallets
• Malware also enables surveillance and remote control; campaign peaked in March 2026 and remains active against Roblox’s 82M players

Cybercriminals are targeting Roblox players with an infostealer and a Remote Access Trojan (RAT) malware that grants them full control over compromised computers, experts have warned.

Roblox is an online gaming platform, virtual universe, and game creation system where users play millions of games and user-generated mods. Among the mods is Xeno Executor, a utility that allows players to run scripts that automate certain actions or run custom code. Some players use Xeno Executor to run cheats, too.

Since it’s an unofficial script, Roblox does not allow it and blocks it whenever a new version is released. Now, security researchers Bitdefender have reported finding an “undetected” version being promoted on various gaming forums, Discord communities, and similar.

CornFlake and CocoShell

This version is advertised as “invisible” to Roblox’s anti-cheat systems, but in reality, all it does is trigger an infection chain that ends in a Java-based RAT and information stealer.

The malware grabs browser data such as passwords and cookies from some of the most popular browsers (Chrome, Edge, Brave, Opera, Vivaldi), as well as online accounts and payment data (Discord, Roblox, Minecraft, Microsoft Store tokens, and more).

It also steals cryptocurrency wallet data, particularly targeting the Exodus Wallet.

As for surveillance, it can log keys, track mouse movements, grab screenshots, stream whatever is on the desktop, and access the webcam. The crooks are also granted file upload and download, PowerShell command execution, and more.

The campaign was kicked off at the start of the year, reaching its peak in March, it was said. It has now stabilized and is still going relatively strong.

We don’t know exactly how many players fell victim to this campaign, but Roblox is an incredibly popular platform, so it is possible the campaign was rather successful, too.

According to Activeplayer, Roblox currently has more than 82 million active players.