Technology

• Bluesky confirms 24‑hour outage was caused by a DDoS attack on August 17 2026
• Researchers linked it to Iraq‑313 Team, using DiamWall‑based DDoS‑for‑hire infrastructure
• Company upgraded defenses; no details yet on attackers, traffic origin, or user impact

The recent outage on Bluesky was the result of a Distributed Denial of Service (DDoS) attack, the company has confirmed.

Bluesky is a decentralized social media platform which is rather similar to X, since it allows users to post short messages and multimedia. Its key difference is the Authenticated Transfer Protocol (AT Protocol) upon which it was built, and which allows users and developers more control compared to other social networks.

On Sunday, August 16 2026, users started reporting problems accessing Bluesky. On Reddit, users from the US, UK, France, and other countries, said they were having issues loading the Bluesky website and app, or accessing their feeds. A day later, on August 17, Bluesky said it suffered a DDoS attack that lasted roughly 24 hours.

Iranians claim the attack

The company did not say who the attackers were, where the malicious traffic originated from, or if any specific DDoS infrastructure was used in the attack. It also did not say how many people were affected, but stressed that it upgraded its defenses and was continuing to monitor the situation.

At the same time, security researchers took to the IFIN public forum to discuss the attacks, saying they saw The Islamic Cyber Resistance in Iraq-313 Team, an Iran-backed threat actor, take responsibility for the attack, as well as for a similar DDoS strike on GitHub. It sounds plausible, since we’ve seen the 313 Team use DDoS to target similar services in the past, including Spotify and Ubuntu.

Their initial research suggests the crooks used DDoS-for-hire infrastructure that relies on DiamWall which, in turn, seems to be using IP addresses supplied by a China-based reseller. This does not mean the attack traffic came from China, or that Chinese entities were involved in the attack.

Via TechCrunch

• Current iPhones could fall in value by 30% once new models launch
• As such, owners should sell them now if they're planning to upgrade soon
• You may be able to lock in a price now but keep using your phone until the iPhone 18 series lands

If you have an iPhone and are planning to upgrade soon, then it’s worth selling your current handset now rather than waiting until the iPhone 18 series lands — even if you’re hoping to purchase one of these upcoming phones.

This is according to Sam Wilson, Director at Gadget GoGo — a site that buys and sells refurbished tech — who claims that pre-owned prices drop dramatically once new models launch, with prices potentially dropping by up to 30% in a matter of weeks.

The issue is biggest with high-end models, as Wilson explains that “if you’ve got a Pro or Pro Max, waiting until after the launch can be particularly costly. We can see values fall by £150 to £300 in a matter of weeks, so selling slightly earlier can make a significant difference to what you get back.”

But even cheaper and older iPhone models can see their prices fall noticeably once this year's models launch, and while prices typically plummet after the launch, data from previous years suggests a steep fall in value can start already. So, whatever iPhone you have, it’s worth selling it now if you’re planning to get rid of it soon.

Lock in that price

The iPhone 17 Pro Max could soon lose a lot of value (Image credit: Lance Ulanoff / Future)

Of course, selling now can present a problem if the phone you want to upgrade to isn’t out yet, but as Wilson explains, you should "start the selling process now, even if you plan to upgrade in September. A lot of buyback services allow you to lock in a price today but keep using your phone for a few more weeks."

And this phenomenon isn’t exclusive to iPhones — other brands reportedly see pre-owned prices fall once new models launch too. So if you have a Samsung Galaxy S26, for example, you might want to sell it before the Samsung Galaxy S27 comes out.

But Samsung’s next flagship phone isn’t expected for a while, whereas new iPhones — likely including the iPhone 18 Pro, iPhone 18 Pro Max, and iPhone Ultra — will probably launch in early to mid-September, giving you less than 30 days before your current handset’s value could fall substantially. Don't say we didn't warn you!

Every year, marketing teams are expected to deliver more. Marketing leaders are under growing pressure to prove the return on investment they deliver and demonstrate the value they bring to the business.

At the same time, budgets remain tight, sales cycles are getting longer, and more people are becoming involved in buying decisions.

Against that backdrop, many organizations have turned to AI to automate routine tasks, analyze market and audience data and support content creation.

But as adoption matures, the conversation is now moving from experimentation to execution.

According to recent UK government research, whilst 41% of businesses are now using AI technology, only 21% have integrated AI into their existing business systems, showing that while there’s been real progress, we're still in the early stages.

The next challenge is not over-use, but shallow use. That means moving beyond isolated use cases and embedding AI into everyday marketing workflows, equipping people with the skills to use it effectively and focusing on the commercial outcomes it helps deliver.

An important part of this is ensuring humans remain at the core. Ultimately, AI should enhance human capability, not the other way around. This is particularly important when it comes to the creativity, storytelling and decision-making at the heart of marketing.

Looking beyond isolated use cases

The shift from AI experimentation to execution requires organizations to look beyond familiar use cases and think more strategically about where AI can create commercial value.

Rather than treating AI as a standalone capability to simply improve efficiency, marketers should be looking at where it can support existing workflows across the entire marketing function - from campaign planning and content creation through to new lead generation, sales enablement, measurement and reporting.

One of the biggest barriers is that people tend to rely on what they already know. That's why ongoing training and development are so important. Helping marketing teams understand where AI genuinely adds value and giving them the confidence to apply it in new ways are what will allow organizations to move beyond isolated use cases and deliver stronger commercial outcomes.

Execution for commercial outcomes

As AI becomes embedded across the marketing function, success will depend less on adopting the technology itself and more on how effectively teams apply it to core business objectives. While use cases continue to evolve, marketing leaders should focus on three priorities to deliver the greatest impact.

First, AI should drive efficiency with a goal to shorten sales cycles. It should be viewed as a tool that removes friction from the customer journey to accelerate the flow of opportunities through the funnel.

Beyond just task automation, marketing teams should be focusing on how their use of AI is delivering faster decision-making, more relevant customer engagement and stronger alignment between marketing and sales - ultimately helping organizations reduce sales cycle length and bring revenue forward.

Second, as AI-powered search experiences increasingly influence how buyers discover information, producing more content is no longer enough. Brands must focus on creating high-quality, differentiated and authoritative content that AI systems are more likely to surface and reference. This means going a step further than using prompts in chatbots and using AI to uncover audience insights, scale ideation and optimize formats.

Finally, marketing teams should leverage AI’s ability to collect actionable insights from large volumes of data and predictive analytics into emerging trends to streamline reporting and measurement processes. Automated dashboards and lead scoring can provide near real-time visibility into performance, enabling leaders to make faster, more informed decisions and keep teams aligned against shared objectives.

In an environment where budgets are under constant scrutiny, robust measurement frameworks will be critical to demonstrating marketing's contribution to business outcomes.

AI should enhance human capability, not the other way around

It is important to remember that human creativity remains essential for developing original perspectives, compelling storytelling and genuine brand distinction. Organizations that combine AI efficiency with human expertise will be best positioned to maintain visibility and relevance in an increasingly AI-mediated digital landscape.

This starts with ongoing training and development, and creating an environment where employees feel confident experimenting, sharing knowledge and executing new capabilities. As AI becomes embedded across the entire workflow, strong leadership and collaboration across marketing, sales and the wider business will be just as important as the technology itself, helping teams apply AI consistently, securely and strategically.

The organizations seeing the greatest success will be those that build AI around their people, rather than expecting people to adapt to AI. When organizations combine the right technology with ongoing training, cross-functional collaboration and clear governance, AI becomes much more than an efficiency tool. It becomes something that enables employees to apply greater creativity, judgement and commercial thinking, while ensuring implementation remains responsible and aligned to broader business goals.

Looking ahead

The conversation around AI has moved past experimentation and is now about getting the balance right - moving beyond shallow, task-based use of AI and embedding it strategically across marketing in ways that deliver measurable commercial value, without overusing it or losing the human qualities that customers value most.

Looking ahead, UK Government research shows that 59% of organizations planning further AI investment intend to embed AI into their existing tools and systems over the next 12 months, while 42% also plan to invest in workforce reskilling and development.

Ultimately, the organizations that pull ahead won't necessarily be those investing in the most technology, but those that invest in their people. This means helping teams understand where AI genuinely adds value, and ensuring it compliments commercial thinking, continuous learning and human creativity, rather than replaces it.

We've featured the best email marketing platforms.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

Adding a lot of dependable capacity to a NAS device isn't cheap, but shopping around can net you major savings. The 8TB Seagate IronWolf is currently $350 (was $518) at Best Buy. But I wouldn't buy it there. Newegg has the exact same drive for $330, saving you a further $20.

IronWolf is purpose-built for NAS use, with 24/7 operation and multi-drive environments in mind. This 8TB model uses CMR recording technology, spins at 7200 RPM and has a 256MB cache, making it a solid choice for expanding a home NAS, small-business server or RAID array.

Seagate rates it for workloads of 180TB per year and 1 million hours MTBF. AgileArray provides RAID optimization and dual-plane balancing. Rotational vibration sensors are designed to help maintain consistent performance when several drives are operating together.

8TB Ironwolf NAS Internal Hard Drive: $329.99
The Seagate IronWolf ST8000VN004 provides 8TB of CMR capacity, 7200 RPM performance, 256MB cache and SATA 6.0Gb/s connectivity, with a 180TB/year workload rating and rotational vibration sensors.View Deal

8TB Ironwolf NAS Internal Hard Drive: was $517.65 now $349.99
This is the exact same drive, and while it's a solid $167.66 off the list price at Best Buy, it's still $20 more than you can pick it up for at Newegg.View Deal

Why we recommend it

In his 4-star review of the 8TB NAS drive, our expert Mark said: "IronWolf drives feature CMR technology and AgileArray firmware, ensuring smooth RAID performance, reduced vibration, and efficient power management. The 180TB per year workload rating, the RV sensors, the three-year warranty, and the bundled Rescue Data Recovery Services have all persisted as defining features of the consumer tier."

The 7200 RPM spindle speed and 256MB cache and SATA 6.0Gb/s interface makes this a performance-focused mechanical drive rather than simply somewhere to dump 8TB of rarely accessed files.

For more options, take a look at our roundups of the best desktop and laptop hard drives and best NAS hard drives.

Price context & historical value

Best Buy is currently selling the 8TB Seagate IronWolf for $349.99, which is a massive $167.66 reduction off the usual $517.65 asking price.

While that looks like a brilliant deal on paper, the identical drive is $329.99 at Newegg, a saving of $187.66 which means you're paying about $41.25 per terabyte.

Should you buy it?

✅ Buy the 8TB Seagate IronWolf NAS drive if...

You need a high-capacity drive to cope with the demands of an always-on NAS. The combination of 8TB capacity, CMR recording and 7200 RPM makes this a strong option for backups, media libraries, and shared files.

It's particularly well suited to multi-bay and RAID setups. AgileArray is designed to optimize RAID performance, while the built-in rotational vibration sensors help the drive maintain consistent operation when it's surrounded by other spinning hard drives.

❌ Skip the 8TB Seagate IronWolf NAS drive if...

You only need additional storage for an ordinary desktop PC. Although you can use an IronWolf as a conventional SATA hard drive, much of what you're paying for is specifically intended for NAS environments, including 24/7 operation, RAID optimization and vibration management.

I'd also look elsewhere if silence and low power consumption are priorities. A 7200 RPM mechanical drive is designed with performance in mind and won't offer the near-silent operation of an SSD.

The Catch: What to know before you buy

NAS manufacturers maintain compatibility lists for their enclosures, and checking your system supports this exact ST8000VN004 model can save headaches later, particularly if your NAS is a little old.

It's also important to remember that RAID isn't a substitute for a separate backup. The 8TB IronWolf can provide plenty of capacity and is designed for RAID environments, but drive redundancy doesn't protect you from accidental deletion, malware, theft or failure of the NAS itself.

Also consider

12TB Ironwolf NAS: was $455.99 now $429.99
The larger 12TB IronWolf model combines 7200 RPM performance with a 256MB cache, up to 210MB/s read and write speeds, IronWolf Health Management, and included Data Recovery Services.View Deal

Motorways don't stay safe simply because drivers follow the Highway Code. They rely on speed limits, lane markings, and crash barriers that keep traffic moving to help prevent mistakes from becoming serious accidents.

AI governance works the same way. Policies define the rules, but operational controls keep organizations on course when technology, users, and risks change.

As organizations accelerate AI adoption, many have focused on developing policies, governance frameworks, and acceptable use guidelines. Those are essential first steps. But policies alone can't prevent AI systems from accessing the wrong data, interacting with the wrong applications, or creating unintended operational risks.

The EU AI Act has made AI governance a board-level issue, requiring organizations that use or develop AI to identify, manage, and document risks, particularly for high-risk systems.

This has introduced new expectations around oversight, documentation, data governance, and transparency. However, regulation alone won't solve the biggest practical challenge: AI adoption is moving faster than governance.

Employees are using public AI tools, developers are relying on coding assistants, and business teams are deploying AI capabilities across existing platforms. In many cases, these technologies are introduced before security, compliance, and risk teams fully understand what they can access, which systems they influence, or how they change the organization's overall risk profile.

That's the gap between regulatory ambition and operational reality. Organizations may have approved AI use cases and governance frameworks, but they're often governing only part of the AI already operating across the business.

Closing that gap starts with treating AI as an operational challenge rather than simply a compliance exercise. In practice, moving beyond policy means understanding how AI is used, applying controls according to business risk, and adapting governance as AI capabilities evolve.

Shadow AI is changing the governance challenge

Traditional governance assumes organizations know what technologies are operating across the business. AI has changed that. Shadow AI is no longer limited to employees experimenting with public tools.

It also includes embedded AI capabilities introduced through routine software updates, AI-powered workflows created by business users, and autonomous agents operating across enterprise systems without ever being treated as formal AI projects. These capabilities are often adopted faster than governance processes can identify, assess, and manage them.

Today, AI isn't simply supporting employees - it's acting on their behalf. AI agents can retrieve information, trigger workflows, interact with enterprise systems, and make operational decisions using permissions inherited from the people or systems they represent.

Because of this, the enterprise attack surface has fundamentally changed. Security teams are no longer protecting only people, devices, and applications - they're also securing autonomous systems capable of moving across environments, accessing sensitive information, interacting with critical workflows, and making decisions independently.

Every new AI agent represents another potential pathway for accidental misuse or malicious activity if it isn't governed appropriately.

Regulation defines the rules. Context builds the rails.

The EU AI Act places significant emphasis on accountability, oversight, and risk management throughout the AI lifecycle. But meeting those expectations requires far more than documentation.

Regulation defines the outcomes organizations must achieve; the challenge is translating those requirements into operational governance across environments where AI is constantly evolving.

That starts with understanding AI in context. Risk can't be assessed by looking at an AI system in isolation. An AI assistant summarizing internal documents may appear relatively low risk until it gains access to commercially sensitive information.

Likewise, an AI agent automating procurement may operate safely under normal circumstances, but its risk profile changes significantly if it can approve financial transactions, inherit privileged identities, or initiate actions across critical business systems.

Risk is defined not only by the AI itself, but by the identities, applications, infrastructure, data, and business processes it connects to. As it becomes more deeply embedded across these interconnected systems, AI governance increasingly overlaps with cybersecurity.

Every AI capability introduced into the enterprise can create new attack paths, expand access to sensitive information, and increase the number of non-human identities operating across the environment.

Recent incidents, such as PocketOS’s database being wiped by an AI coding agent in seconds, have shown how quickly AI risk can become operational risk when AI systems are granted more authority than intended.

Organizations should therefore govern AI identities and permissions with the same rigor as human users, regularly reviewing inherited access and removing unnecessary privileges before they create new attack paths.

Governance also needs to extend beyond individual AI systems. Organizations should understand how AI interacts with identities, enterprise assets, business processes, and critical data so they can prioritize controls where they'll have the greatest impact.

Bringing security, risk, compliance, and technology teams together around this shared understanding turns governance into an enabler of resilience rather than simply a compliance exercise.

Building the rails

As AI regulation continues to evolve, organizations will be expected to demonstrate that governance exists not only on paper, but in practice. Policies establish the rules, but operational governance builds the rails that keep AI operating safely as technologies evolve, permissions change, and new capabilities emerge.

Moving beyond policy means embedding governance into day-to-day operations. Organizations should continuously identify new AI capabilities, understand how they relate to enterprise assets, review the permissions they inherit, assess business impact, and prioritize the exposures that present the greatest risk.

This allows teams to apply controls where they'll have the greatest impact while supporting responsible AI adoption across the business.

Governance also needs to keep pace with change. As AI agents evolve and environments shift, organizations should be able to adjust permissions, restrict access, isolate affected agents, or prevent actions that exceed an AI system's intended role before they become wider operational issues.

The objective isn't to slow AI adoption, but to create the confidence to innovate securely as AI becomes part of everyday business.

Innovation needs both rules and rails

AI will continue to transform how organizations operate, innovate, and compete. But organizations that realize AI's full potential won't be those with the longest policy documents.

They'll be the ones that embed governance into everyday operations, govern AI identities and permissions with the same discipline as any other critical asset, and continuously adapt controls as technology evolves.

The EU AI Act has raised expectations around accountability, but organizations that move beyond policy and compliance will be best placed to unlock AI’s full potential.

In the age of autonomous AI, governance isn't just about setting the rules - it's about creating the guardrails that let organizations innovate with confidence without compromising security, trust, or resilience.

We've featured the best AI website builder.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Microsoft Teams is changing how it displays swear words
• The toggle to hide bad words is now Off by default
• However Government users will have to wait a little longer, for an unknown reason

Frustrated Microsoft Teams users finally be able to express their true feelings after the company revealed changes to how it censors profanity in captions - although not all users will be able to "benefit" just yet.

In an updated message center post, Microsoft outlined the changes coming to its profanity filter for Teams, noting that the default setting for live captions has now been switched from On to Off.

However this doesn't yet apply to all users, as Government officials and staff will have to wait a little while longer, with their rollout deadline now extended until sometime at the end of August 2026.

Safe for work

Microsoft first introduced the profanity toggle in 2023 in a bid to cut down on bad language in Teams Live Captions by replacing any swear words with asterisks.

Originally, it was up to the user whether to leave the toggle on, rather than the meeting organizer or admin - but in May 2026, Microsoft appears to have changed its mind for some reason, and opted to disable the "profane words" filter by default.

"This change ensures captions more accurately reflect spoken content, supporting accessibility needs and aligning with regional regulatory requirements (including the EU)," the company wrote.

Microsoft added that users retain full control and can enable or disable the filter at any time, but, somewhat tongue in cheek, notes that, "No admin action is required, but informing support teams and users is recommended."

The delay will apply to Government Community Cloud (GCC), GCC High, and US Department of Defense users, who often face extra complications due to security checks and operational reasons - although Microsoft offered no clues as to why that is the case here.

The news is the latest in a series of recent updates seemingly geared towards helping Microsoft Teams users show off how great they are.

This includes bringing "Badges" to your Microsoft Teams profile card, allowing users to display all their achievements to anyone who cares to look.

It is also working on a feature which will automatically update a user's work location when they connect to an office Wi-Fi network - hopefully meaning less confusion about where workers actually are situated, but bad news for those looking for a quiet day in the office tucked into a corner.

Via The Register

• Amazon has rolled out tougher DRM to the Kindle Oasis
• The same update has previously landed on most other Kindle models
• The company has also closed an airplane mode loophole for Libby loans

If you like reading Kindle books on other devices or borrowing digital library books using Libby, Amazon has just made some changes that you’re unlikely to be happy about.

First up, as reported by The eBook Reader, a software update has rolled out for the Kindle Oasis that adds a new hardened DRM (Digital Rights Management) to books on it, meaning that most DRM-removal tools will no longer work.

While we don’t condone piracy, there are legitimate reasons to want to remove DRM, such as reading the book on a non-Amazon ereader that doesn’t have access to the Kindle app.

And this isn’t the first or only Kindle to get this undesirable update — now, the only supported model that doesn’t have this new DRM is the Kindle Voyage, and it’s probably only a matter of time before that gets it too.

The Amazon Kindle Oasis (Image credit: TechRadar)No more lengthy loans

The other change that Kindle users are unhappy with is that you can seemingly no longer keep a Libby book beyond its due date by turning airplane mode on. This was a handy workaround that gave you more time with a borrowed book, but now Reddit users report being prompted to turn Wi-Fi on to access their loans.

As these are digital loans, this workaround wasn’t really hurting anyone — the books would still be made available to other users after the loan period expired anyway; you’d just also still have a copy on your Kindle.

Unsurprisingly, the online reactions to these changes aren’t positive, with Reddit comments about the new DRM including the likes of “glad I switched to Kobo,” while of the Libby change, one commenter claimed it was “my biggest nightmare.”

If you’re affected by either of these issues, then there aren’t any easy solutions — other than potentially switching to another brand. In the case of the DRM, Kobo books currently use a version that’s easier to remove, but even on Kobo, the airplane mode loophole won’t work.

• Technics celebrates 55 years of its hugely popular SL-1200 turntable
• New anniversary edition teased, full launch in September
• Adds gold, gold, and more gold to the existing model

Since the launch of the original Technics SL-1200 in late 1972, the model has proven a staple of hip-hop and DJ culture, and newer editions such as the SL-12000GR2 keep winning us over. With a milestone coming up, the brand is set to celebrate its birthday in style.

Technics has unveiled a 55th anniversary model of the SL-1200 (despite it only being 54 years since 1972, but who doesn't fudge the numbers when a birthday is coming up?) which is set to be fully unveiled in September.

We've had a good look at the new record player already, thanks to an early tease from the brand, so we know what it's all about. The only actual information we still need is how much the SL-1200 55th Anniversary Edition will cost — but given the design, there might be a surcharge to pay over the closest original.

The current MK7 model goes for $1,200 / £849 / AU$1,699, and since that's the platform that the Anniversary Edition is based on, it'll likely cost a fair bit more.

How to celebrate a birthday

The main distinguisher for the SL-1200 55th Anniversary Edition is the addition of lots of bling.

Tonearm base? Gold. Knobs and buttons? Gold. Platter? Gold rimmed. Tonearm? Gold-anodised. Don't forget the gold slip mat. And, as you can see at the top of this article, there's a fully-gold color model you can buy. However there are also blue, black, white, red, green and gray models, which give the gold something to actually contrast.

For clarity, I'm not knocking it: celebrate your birthday however you want, Technics.

Features and tech-wise, Technics has made a effort not to play about too much with what makes the SL-1200 work. If it ain't broke, don't fix it. It's really the gold trim which makes this stand out.

As mentioned, it's based on the SL-1200 MK7. This means it's a manual turntable with an aluminum platter, stylus-illuminator LED, Universal Static Balance tonearm, pitch adjustment dial and phono output.

Now, turntable fans might be getting deja vu here. The brand is no stranger to fancy editions, even gold-flecked ones, releasing a SL-1200 Master Edition last year with perhaps 55% as much gold as the new birthday model.

Thermal Master DV2: 30-second review

The Thermal Master DV2 is a handheld thermal imaging monocular that the maker positions in different ways depending on which storefront is visited. On its own birdwatching page, it is billed as the first dedicated thermal camera built specifically for spotting birds and wildlife.

Elsewhere in the same company's catalog, and on Amazon, it appears instead as a general-purpose thermal monocular for hunting, camping and rescue work. Whatever your interest, the hardware underneath is identical.

At the centre of the device sits a 256 x 192 sensor with a 12-micron pixel pitch. Thermal Master runs that raw feed through what it calls the RazorX engine, a three-core processing pipeline built from an FPGA, an ASIC and an AI upscaling block badged X³IR. The result is a 512 x 384 output image, twice the native resolution, though this is achieved through interpolation rather than by a genuinely higher-resolution sensor.

Detection is rated up to 1000m for a human-scale target, with shorter recognition distances quoted for smaller subjects such as birds. A 10mm f/0.8 manual focus lens feeds that sensor, and the whole thing is viewed on a 5-inch touchscreen running at 1280 x 720. That screen rotates and folds, making it compact to carry and easy to use. And the camera can be used without a phone, making deployment quicker and easier.

Connectivity can be used for streaming over Wi-Fi or for offloading captured images and video from the camera, and it supports RTSP for near-no-lag transmission. A laser pointer helps mark whatever has been spotted, and a Picatinny rail fitting lets you mount the DV2 on other equipment.

Thermal Master sells the DV2 directly from its own site and through Amazon in several regions. Pricing is $499/£459 via Amazon, which seems reasonable since this isn’t a phone add-on, but a standalone camera.

The value of this product entirely pivots on the use case. If you need to identify animals or people at range, then it could be invaluable, but it's not suitable for every job.

(Image credit: Mark Pickavance)Thermal Master DV2: price and availability

• How much does it cost? $500/£460/€530
• When is it out? It is available now
• Where can you get it? Direct from the Thermal Master website, with shipping from regional warehouses in the US, UK, EU, and China. And via online retailers.

At the time of review, the DV2 carries a price of $489 USD direct from Thermal Master. For that price, you get the camera, a battery charger, one battery and a few accessories, like the Picatinny rail attachment and carry case.

For quick delivery, it can be purchased from Amazon.com for $499, from Amazon.co.uk for £459.99 and from the EU Amazon sites for €529.99.

There aren’t any obvious retail competitors because professional bird watching isn’t something most companies see as a new market opportunity.

A more general-purpose device is the Axion Compact Thermal Monocular, but that’s priced at $1449.97 on Amazon for those in the USA.

A slightly more realistic choice is the TOPDON TS004 and TS004 Pro, monoculars designed for hunting, rather than birdwatching. The standard TS004 has a resolution of 256 x 192 and costs a little over $400, while its Pro brother is $759.95 and sports 384 x 288 resolution. The Pro quotes detection at up to 650 yards, which is less than the DV2.

(Image credit: Mark Pickavance)

• Value score: 3.5/5

Thermal Master DV2: Specs

Spec

Detail

Sensor

256 x 192, 12μm pitch

Sensitivity (NETD)

≤20mK @ 25°C, f/0.8

Super resolution output

512 x 384 via X³IR interpolation

Lens

10mm f/0.8, manual focus

Field of view

17.5° x 13.1° (some listings simplify this to 18°)

Display

5 inch touchscreen, 1280 x 720, 360° rotating, 180° folding

Refresh rate

50Hz

Processor

Triple-core: FPGA, ASIC, AI upscaling block (RazorX engine)

Detection range

Up to 1094 yards for a human-sized target

Digital zoom

8x

Reticles

9 tactical modes, 5 colours

Colour palettes

5

Connectivity

Wi-Fi streaming, RTSP support confirmed independently, USB-C

Storage

32GB internal

Mounting

Picatinny rail adapter

Weight

512g including battery

Battery

1x 21700 5000mAh Battery

Warranty

2 years

Thermal Master DV2: Design

• Camcorder vibe
• Large touch screen
• Only one battery included
• Confusing buttons

Many handheld thermal cameras still borrow the shape of a barcode scanner or a torch, but the DV2 looks closer to a small camcorder. The optics are at the top end, with a lens cap protecting the manual-focus telephoto. The lens cap isn’t tethered, so I recommend not dropping that black item in the dark the first time you use this device.

Its defining feature is the fold-out 5-inch touchscreen, which rotates 360 degrees and folds 180 degrees, rather than staying fixed in one position. This lets the operator adjust the display if the target area is above or below the camera's natural line of sight.

The rotating, folding screen lets the display angle toward the eye, whether the DV2 is held low, mounted on a tripod, or fitted to a rail system via its Picatinny adapter.

A downside of this design choice is that if you fold out the display at 90 degrees to the camera, its most natural position, the weight causes the camera to fall over. Therefore, at a minimum, you need a small tripod, even if one isn’t included with the DV2.

To operate the device, the red button on the rear is depressed for what seems like an inordinately long time before a fancy video graphic announces that the DV2 is ready for action.

But before you do that, the 21700 battery needs to be inserted into the DV2, something you are told to do, but there are no indicators I could see which end of the battery goes in first.

Eventually, I found a video of someone demonstrating the camera and realised the correct orientation, but it wasn’t labelled or mentioned in the quick start guide.

Documentation for this product isn't a strong point, since months after release there is still no proper manual.

(Image credit: Mark Pickavance)

What’s also slightly odd about the battery is that the case has positions for two, but Thermal Master provides only one. And yet, the charger accepts two. Given that the retail unit cost of a 21700 battery is less than $10, and it would be much less than that for Thermal Master, it's confusing why two aren’t provided from the outset.

If you don’t intend to buy another, then the charger is also superfluous, since you can charge the one included in the DV2 by using a provided USB-C cable. Except you might not want to, because you might struggle to power the DV2 while it's on a tripod, since the USB-C port is on the same underside as the threaded tripod hole.

That’s confusing, but the bit that genuinely made me scratch my head was that the second button next to the power on/off one didn’t initiate video and picture capture. No, it toggles through the different colour representations, something most people would set and forget.

To take a picture or video, you have to touch the screen, which, if you're zoomed in on a distant bird or animal, is likely to twist the camera and ruin the framing. Not sure who made the choice not to include a shooting trigger, but it was a poor decision.

Another slightly odd feature is that this device has a laser pointer, activated by a quick press of the power button. I can’t think of anything more likely to cause roosting birds or other wildlife to exit stage left than being shot in the eye with a laser, so why they included this, I’m unsure.

I'm also confused by the addition of the Picatinny rail adapter, which implies that after you've watched the birds, your next step is to shoot them. That's not really the sort of bird watching I'm familiar with, but what do I know?

Overall, the build quality is decent, and it feels relatively robust. From a design perspective, there appear to be numerous aspects where the engineers creating it didn’t really appreciate how it might be used and what potential customers might be expecting.

(Image credit: Mark Pickavance)

Design score: 3.5/5

Thermal Master DV2: User experience

• Questionable ergonomics
• Dire mobile app
• Focus challenges
• Recognition vs detection

I might have got the wrong end of the stick, but the primary purpose of this device is to take pictures and video using the thermal sensor, isn’t it?

This seems obvious, but due to how poorly organized the device menus are for doing just that, perhaps I’ve entirely misunderstood this device.

I’ve already mentioned that you can’t initiate capture with a simple button, but the touch screen menu goes one step further. Taking a picture isn’t on the top-level menu; it requires you to go to a second page, and then a third, to do the one task this device is made for.

The top level is full of things you won’t want immediate access to, I’d contest, and this arrangement makes almost no sense I can fathom.

At first, I concluded you were probably meant to use the mobile app to take pictures so you don’t touch the camera, but I was wrong.

Connecting the camera to the phone involves an overly complicated procedure that would make anyone used to Apple or Bambu Labs products incandescent.

For starters, both devices need to be on the same Wi-Fi network, so this won’t work in the field unless you bring a portable Wi-Fi access point along.

The phone needs an app that, for Android users, needs to be downloaded from the support pages and then side-loaded onto the phone. I’ve long been critical of Thermal Master apps, and how there is always a new one for each camera, and this one isn’t even on the Play Store.

Once you install the app, you can go to a specific page on the camera with no instructions on which icon it's behind, then get a QR code that connects the two.

To use this feature, you must be in control mode, where the barcode is shown, so you can’t do anything on the camera when the phone takes over. Once the two are connected, you have three features.

One of which is the color control, another is the Thermal Master eye (Red/Green imaging mode), and the other recalibrates on the image. There isn’t any way to change from still image to video, use the aiming reticule, hotspot tracking, or access the picture album, so 80% of what the DV2 does is unavailable.

But also, you can’t take a picture other than using your phone's inbuilt feature to take a screenshot. And, if you capture images using the phone, they are not transferred to the phone.

So anything you manage to capture needs to be downloaded from the camera in a separate exercise using a USB-C cable.

Being brutally honest, it appears that the hardware team started work on the DV2 many months before the software people, who appear to have met it a couple of weeks before it was launched.

I suspect that this connectivity was supposed to use Bluetooth, but time ran out, and it also did for all the other missing features of the phone app.

If you ignore the phone app nonsense - and I recommend you do - and focus instead on the functionality built into the camera, then the user experience is much better, if a little limited compared with some of the other Thermal cameras that this brand makes.

What’s lacking here is a visible-light sensor, the data from which could be used to enhance thermal captures and help with focusing. I’d put focusing as the single most challenging part of using the DV2, because when you're presented with blobs of moving colour on the screen, it's hard to tell whether they’re moving in or out of focus.

Once focus is dialled in, as long as you don’t move the camera, you definitely need a tripod; then you can develop a practical workflow.

(Image credit: Mark Pickavance)

What’s important to realise is that this is for capturing things that aren’t especially close, and this hardware would be of zero use to anyone repairing electronics or fixing engine problems. And, because it's not meant for that, some of the features seen on other cameras, like floating temperature text, are missing on the DV2.

Also, the range of this device allows you to identify the thermal signatures of humans and animals long before you can determine where they are going or what they are doing.

Hilariously, to make this understandable, the DV2 datasheet breaks this down into four detection range levels, where the thing being detected will trigger up to three pixels on the sensor. These are 300m for a 1.8m human, 250m for 1 1.5m Hog, 200m for a 1.2m Coyote, and 50m for a 0.3m Dove.

Where I live in England, we have only two of those creatures since Wild Hogs and Coyotes aren’t endemic, but I get the idea. The camera's relatively high frame rate helps with identification, but only when they’re larger than three pixels on the sensor.

Thermal Master marketing takes this capability and extrapolates it to claim the camera can detect human-sized heat signatures at 1094 yards (1000m, or 1km). To be clear, at that range, an adult human would be a single pixel on the sensor, and for them to be recognised as such, they’d need to be four times closer.

I guess the single pixel, if consistently lit, might tell you something is coming, just not what it was.

(Image credit: Mark Pickavance)

• User experience score: 2/5

Thermal Master DV2: Camera

(Image credit: Mark Pickavance)

Image output is a 512 x 384 frame, built from a native 256 x 192 sensor through the RazorX engine's AI upscaling. As with any interpolated output, this can improve apparent detail without genuinely increasing sensor resolution.

And, in the DV2, this works well enough, making the thermal image clear and readable, with good separation between different heat sources even at a distance.

Digital zoom is available up to 8x, but zooming this way involves taking smaller and smaller sections of the sensor area, which results in grainy/blocky results. When you start with such a low sensor resolution, this is always going to be an issue.

(Image credit: Mark Pickavance)

Five colour palettes are available for translating temperature data into a viewable image, along with nine tactical reticule modes in five colours, a feature set that leans towards the hunting and outdoor positioning of the device rather than its birdwatching marketing. Since that type of hunting isn’t common in the UK, I can’t comment on its usefulness for that role.

But in other regions, where the number of people accidentally shot by others during the hunting seasons is significant, it could be a lifesaver.

A 50Hz refresh rate should keep motion reasonably smooth when tracking a moving subject, whether that is a bird taking flight or an animal moving through undergrowth. The issue here is that there is a sweet range where this can happen, but the further out a target is, the more difficult it becomes to track, especially if it moves out of frame.

There is a tendency to scan wildly around, trying to locate the heat source before it moves behind an obstruction or leaves the area entirely.

The camera offers similar technology to what Thermal Master is providing with its accessory phone cameras, but the optics here are predictably better. That results in better end results and clearer images, even if the resolution is still relatively low.

• Camera score: 3/5

Thermal Master DV2: Final verdict

(Image credit: Mark Pickavance)

The device's underlying hardware is solid. However, other aspects of this device are less than perfect, and the lack of proper documentation alongside some less-than-polished software detracts from the overall experience.

That’s a shame because the fundamental foundations of this device are good; it just needed to come to market in a more holistic package than it did.

I hope Thermal Master is motivated to finish the DV2 like it patently should have been. With some effort on the interface and software components, it could be something all dedicated birdwatchers and wildlife enthusiasts will want in their backpacks.

But at this time, especially in terms of the software and menu ergonomics, it still needs some work.

Should I buy a Thermal Master DV2?Thermal Master DV2 Score Card

Attributes

Notes

Rating

Value

Not inexpensive, but impressive specifications

3.5/5

Design

Poor button choices and menu arrangement

3.5/5

User experience

Limited phone app that works over WiFi

2/5

Camera

Zoom is digital, and sensor is 256 x 192, but the optics help

3/5

Overall

Solid hardware let down by rushed software

3.5/5

Buy it if...

You are tracking at range
The ability to pick up thermal data at a range of 1km is useful, but only if you are looking for people or larger animals. This camera can determine people-sized heat signatures at range, rather than small birds at a distance.

You wish to remain unseen
A handheld thermal monocular with a rotating, folding touchscreen rather than a fixed eyepiece is an advantage if you wish to remain out of sight. Light from the screen might be visible at night, but you can adjust the brightness.

Don't buy it if...

You want to capture using a phone
At this time, the phone app can relay a live feed using RTSP, but it can’t capture images or video. And, the connection between the two requires a WiFi network. Those expecting complete remote control with a phone will be disappointed.

Your interest is birds at range
If small bird detection at real distance is your primary use case, then this is a problem. Because the sensor is low resolution and there is no optical zoom.

Before I watched any of Lioness season 3, I really believed that Joe (Zoe Saldaña) was invincible. She's cheated logic and death multiple times over across the first two seasons, but this Russian intelligence plot just might get the best of her.

This season's drama revolves around two simultaneous timelines: one, Joe being kidnapped by unknown Russian operatives and, two, the six months leading up to her capture. Why kidnap? Because the CIA captured one of Russia's own after foreign intelligence tipped them off.

As of writing, we still don't know if she's made it out the over end unscathed. So, when does Lioness season 3 episode 4 arrive on Paramount+?

What time can I watch Lioness season 3 episode 4 on Paramount+?

Lioness season 3 episode 4 will drop on one of the world's best streaming services in the US and Canada on Sunday, August 23 at 12am PT / 3am ET.

Here's when it will be released in other nations globally:

• US – 12am PT / 3am ET
• Canada – 12am PT / 3am ET
• UK – 8am BST
• India – 12:30pm IST
• Singapore – 3pm SGT
• Australia – 5pm AEST
• New Zealand – 7pm NZST

When do new episodes of Lioness season 3 come out?

(Image credit: Paramount+)

Lioness season 3 will have a total of eight episodes, with new entries airing weekly. That gives us the following schedule:

• Episode 1: out now
• Episode 2: out now
• Episode 3: out now
• Episode 4: August 23
• Episode 5: August 30
• Episode 6: September 6
• Episode 7: September 13
• Episode 8: September 20

Across the UK, "will AI take my job?" has become one of the nation's most searched fears, a concern echoed in a recent open letter signed by more than 200 economists and technology leaders, including former Google CEO Eric Schmidt.

The letter warned that without the right guardrails, increasingly powerful AI could drive large-scale job displacement.

This anxiety around job stability has, unsurprisingly, increased in line with some of the world's largest employers - including Microsoft and Meta - announcing cuts to thousands of roles over the last year.

Left to interpret what that means for their own futures, employees globally have become accustomed to fear and uncertainty.

The impact is far-reaching, with recent research indicating that widespread anxiety is now impacting career decisions. In fact, as many as one in three workers say they are planning to retire earlier than planned, while 25% of workers are considering moving into industries less exposed to AI.

Although concerning, this potential “white-collar exodus” isn't a rejection of AI itself; most workers say the technology makes them productive when introduced with proper training and clear goals. However, it’s a response to how AI is being rolled out: often with little clarity, minimal support and not enough acknowledgement of the potential impact on employees’ confidence in their own expertise.

Left unaddressed, the impact will be felt in retention, engagement and institutional knowledge. Designing and prioritizing ways of working that incorporate transparency and training, while prioritizing job satisfaction and performance, has become a board-level priority.

The cost of the exodus

Many workers feel that AI could reduce the need for their role within a few years, and that fear is not misplaced. Finance, professional services, and IT jobs have already felt this pressure, with PwC cutting 200 entry-level roles this year.

But we must not fall into the trap of reading this as solely a graduate or entry-level problem. In reality, according to a Gartner survey of C-level executives, 56% are extremely likely to quit due to the impact of AI and burnout. This is in some ways more harmful, presenting a clear risk of knowledge attrition across organizations as expertise born of human experience is lost.

Job security writes only the first chapter. Knowledge workers are increasingly frustrated that AI allows tasks once requiring years of expertise to be completed by almost anyone. The concern is about whether experience, judgement and specialist knowledge still carry value. And when organizations fail to redefine what human expertise contributes, disengagement festers.

Compounding this is the growing issue of AI fatigue. More than half of all content online is now AI-generated. So workers clocking off from AI at work are simply wading into more of the same at home, extending far beyond the 9-5 that businesses can measure or manage. Harvard Business Review has linked AI fatigue to decision fatigue, errors, and higher intent to quit.

In response, employees are actively cutting back their use of AI tools. However, investment shows little sign of slowing. Enterprise AI spending reached £1.2 million last year, up 108% on the year before, highlighting the growing gap between organizational investment and employee adoption and engagement.

This combined effect results in a workforce that's disengaging, under-using costly tools and potentially planning their exit. This demands urgent action from businesses, who need to address how AI is implemented and communicated, and act before disengagement turns into departure.

The path forward: Building trust through transparency

Positively, some organizations are already demonstrating what successful AI adoption looks like, grounded in an understanding that communication and culture should be taken as seriously as the technology itself.

IBM, for example, is both expanding and redefining entry-level hiring for an AI-first workplace, detailing where the organization is headed. Since workers often don't understand why they're expected to use AI in their role, moves like IBM’s can help to close a critical communication gap.

Transparency, clear guardrails and genuine dialogue are what turn AI spend into engagement. When people understand where they fit into the future of work, they're far more likely to embrace the technology shaping it.

Reducing digital noise

The instinct in most organizations is to let AI creep into everything at once. In practice, AI delivers the most value - and does the least damage to worker purpose and confidence - when it targets administrative burden and repetitive tasks.

AI tools shouldn’t undermine the judgement, experience and specialist knowledge that people bring to their roles. Careless implementation of AI is the fastest route to the fear of expertise erosion that drives people toward the exit.

Investing in reskilling and upskilling

Adaptavist’s recent research into the human cost of AI reveals that 74% of workers are already taking the initiative to build AI skills, showing that willingness to adapt isn't the problem. But organizations must match that commitment. Rather than treating AI training as a one-off exercise during rollout, businesses should implement continuous learning and support that evolves alongside the technology.

The value of this approach is already being recognized widely. The UK Government has introduced its AI skills program with the ambition of training 10 million workers, backed by partners including the NHS. The recent “skills compact” plan also puts the onus on employers to address skills gaps in AI across financial services.

Both initiatives reflect a growing understanding that successful AI adoption depends as much on investing in people as it does on technology and tools. Effective training should help employees understand how their expertise continues to add value, giving them confidence that AI is there to enhance their role, not to replace it.

Building psychological safety around experimentation

An important reality is that none of the above works if people are afraid to use the tools they're being trained on. Infosys has acknowledged the importance of psychological safety by framing AI adoption as an opportunity to experiment and improve without fear that mistakes will damage confidence or career prospects.

A study published in Occupational Medicine found employees who had 1:1s five times or more within two months reported significantly higher psychological safety than those with fewer meetings. Direct conversations to understand concerns, discuss how AI fits into their career development, and address uncertainty also show a marked improvement.

None of these approaches requires organizations to pare back their AI ambitions. Ultimately, the businesses that retain their best employees will be those that treat AI anxiety as a change management problem, measured by whether their most experienced people still feel essential to the work.

We've listed the best job sites.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

As the resident NBN and home-internet expert for TechRadar Australia, I spend much of my time looking out for the best internet deals, and there are almost invariably a number of compelling options at every different speed tier — and these can be a great way to sense-check if you’re overpaying for your current NBN service.

When choosing deals, much of my decision-making revolves around two easily identifiable metrics: price and typical evening speed (TES). Typical evening speed is something all internet service providers (ISPs) have to disclose before you sign up for a plan. In essence, it tells you the download speed you can realistically expect to achieve during the busier evening hours of 7pm to 11pm. Being able to view the TES also means you can easily identify which ISPs ultimately offer better value — i.e. the ones advertising fast evening speeds at competitively low prices.

With almost all of the best NBN 500 plans promising to deliver their maximum 500Mbps speeds even during the peak evening period, it’s a tier where direct price comparisons make sense. However, things get a bit wobblier when you’re shopping for faster speeds, such as NBN 750 and NBN 1000.

Most plans on these tiers make more conservative promises when it comes to typical evening speed — for example, rather than the 1,000Mbps you might expect from NBN 1000, on these gigabit plans, most providers only promise evening speeds in the 820-860Mbps range. With that TES figure closer to 750Mbps than 1,000Mbps, it prompted me to ask a simple question: Is NBN 1000 actually worth it, or should you save yourself some moolah every month and opt for NBN 750 instead?

Cue a whole load of number crunching!

So average

First, I calculated the current average monthly cost of NBN 750 and NBN 1000 plans, alongside the average TES advertised by the providers I monitor. After crunching the numbers on NBN 750 plans from 14 providers and NBN 1000 plans from 20, the final results were:

Measure

NBN 750

NBN 1000

Average price

AU$105.98

AU$114.95

Average TES

730.2 Mbps

835.4 Mbps

As expected, the results show that NBN 1000 plans do not, on average, get as close to their nominal 1,000 Mbps tier speed as NBN 750 plans get to 750 Mbps. The gigabit gap was quite substantial, falling 164.6Mbps shy of the tier’s nominal 1,000 Mbps tier, compared to a shortfall of just 19.8 Mbps for NBN 750.

With the average monthly cost less than AU$10 difference between the two tiers — specifically AU$8.97, or 8.5% — the question then became whether the NBN 1000’s typical evening speeds are worth it from a value perspective. For that, we of course needed to do some more maths:

Measure

NBN 750

NBN 1000

Shortfall from theoretical maximum

19.8 Mbps

164.6 Mbps

TES as share of maximum

97.36%

83.54%

Cost per advertised TES Mbps

14.5 cents

13.8 cents

Next steps

With NBN 1000 plans falling far short of 1,000Mbps in their typical evening speed promises (averaging just 83.5%), intuitively they seemed like they’d come out second fiddle to NBN 750. Surprisingly, however, NBN 1000 is slightly ahead when comparing cost per megabit for typical evening speeds: 13.8 cents for NBN 1000 versus 14.5 cents for NBN 750. There are a couple of NBN 750 plans that come in below that average at 13 cents per megabit — Kogan Internet and Spintel — which I've listed below.

In other words, NBN 1000 might cost more in absolute terms, but it does offer slightly better value per advertised megabit. If you need some recommendations as to which NBN 1000 plans are worth checking out, look no further than Dodo and Superloop. The cost per megabit on these two works out to be 13 cents, placing them among the cheapest for the tier.

Spintel Home Turbo Fast | 750Mbps | AU$69p/m (first 6 months, then AU$94.95p/m)

If it's the most affordable NBN 750 plan you seek, you simply can't ignore this option from Spintel. It has gone slightly lower in price before, yet it continues to be the cheapest I can currently find in the first year. Spintel is also only one of five providers on this tier to claim maximum plan speeds.

• AU$69 total minimum cost
• AU$983.70 first year cost
• AU$1,139.40 ongoing yearly costView Deal

Kogan Internet Platinum | 740Mbps TES | AU$84.90p/m (first 12 months, then AU$94.90p/m)

The introductory cost of this Kogan Internet plan is evidently much higher than Spintel's, but the telco is currently offering a 12-month discount that results in this being the second-cheapest in the first year. Plus, with the ongoing cost being practically the same as Spintel's, you'll be paying the same 13 cents per megabit.

• AU$84.90 total minimum cost
• AU$1,018.80 first year cost
• AU$1,138.80 ongoing yearly costView Deal

Superloop | 860Mbps TES | AU$84p/m (first 6 months, then AU$114p/m)

Superloop may have increased the price of this gigabit plan as of July 1, but it continues to represent great value. Not only is the TES claim among the highest for the tier, but the telco receives overwhelmingly positive reviews from customers.

• AU$84 minimum cost
• AU$1,188 first year cost
• AU$1,368 ongoing yearly costView Deal

Dodo nbn1000 | 850Mbps TES | AU$77.99p/m (first 6 months, then AU$109.99p/m)

If you want to try NBN 1000 for as little money as possible, this plan from Dodo is tough to beat. You'll pay just AU$77.99p/m for the first six months, which represents the most affordable intro cost for this tier. What's more, with an 850Mbps TES claim, you'll pay the same 13 cents per megabit as the Superloop plan above.

• AU$77.99 minimum cost
• AU$1,127.88 first year cost
• AU$1,319.88 ongoing yearly costView Deal

Is there a third option?

The NBN 750 and 1000 plans don’t exist in a vacuum, however, and NBN Co has been pretty clear that it considers NBN 500 — introduced last September — to be the current sweet spot for most Australian households. So how does the golden child stack up, you ask? I’ve also made it no secret that I consider it the best option for most Australian households, but admittedly, that’s been without these recent calculations.

As we alluded to above, unlike NBN 750 and 1000 plans, most providers do advertise maximum plan speeds for NBN 500, with an average typical evening speed of 499.3 Mbps across 22 providers — equivalent to 99.9% of its potential 500 Mbps maximum.

On a cost-per-megabit basis, however, things don’t look as pretty for my personal fave. With an average monthly price of AU$95.75, the average megabit cost works out to around 19.2 cents, making it substantially more expensive than both NBN 750 and NBN 1000.

With that being said, it still remains the cheapest option in outright monthly terms, with the average NBN 500 plan costing AU$95.75 — AU$10.23 less than NBN 750 and AU$19.20 less than NBN 1000.

As for my top recommendations for an NBN 500 plan, look no further.

Exetel One | 500Mbps TES | AU$80p/m

This Exetel plan has been my favourite ever since it launched in July last year. It offers fuss-free billing, claims maximum plan speeds — the telco has even been found to overdeliver on speeds in ACCC reporting — and it comes with extra perks, giving you the to boost your speed even further for AU$1 a day. The cost per megabit is also the cheapest at 16 cents. It's simply the best.

• AU$80 minimum cost
• AU$960 yearly costView Deal

Kogan Internet Gold Plus | 500Mbps | AU$69.90p/m (first 12 months, then AU$85.90p/m)

If you want to pay as little as possible for an NBN 500 plan in the first year of service, then Kogan Internet's plan is the one for you, thanks to a generous 12-month discount. It also comes in at 17 cents per megabit, well under the current average. Plus, the telco receives generally favourable reviews from customers.

• AU$69.90 minimum cost
• AU$838.80 first year cost
• AU$1,030.80 ongoing yearly costView Deal

Decisions, decisions

So, where to go from here? What these numbers helpfully confirm is that if you do opt for a faster plan, you are getting good value from a purely cost-per-megabit metric — so if you were looking to justify jumping up to NBN 750 or NBN 1000, you can probably stop reading here.

These facts and figures can’t take into account all the subjective-but-important real-world factors that come into play when choosing an internet plan — they can’t, for example, help you determine how much speed your household actually needs.

While it’s now been proven that NBN 1000 plans offer the cheapest cost-per-megabit, that’s only really useful if you actually need gigabit speeds regularly. What you’re ultimately paying extra for, most of the time, is time — specifically, less time waiting for large files (like video games or OS updates) to download.

And with NBN Co recently claiming that average monthly data usage increased to 574 GB per premises — up 13% year-on-year — having a plan that will download all those gigs faster is going to become more and more essential.

For me personally, I’ve decided I can stand to wait a little longer on the one or two occasions a month when that extra speed would be noticeable. It’s hard to justify paying more if the extra megabits are going unused 95% or more of the time.

• Google is building on its new Insulin Resistance Trends health feature with PhotoScan
• PhotoScan is 'an investigational deep learning framework' that can estimate body fat using a phone's camera with 'near DXA accuracy'
• It's not yet available to the public, but the research 'demonstrate[s] the feasibility' of such features

As the Google Pixel Watch 5 launched last week alongside a slew of new Pixel phones, one of the most interesting new developments was the introduction of the Pixel Watch's new Health Guardian suite of tools, including Insulin Resistance Trends.

Insulin Resistance Trends showcases a month-by-month breakdown containing estimations of how your body responded to blood sugar spikes — a crucial component in weight management and the development of Type-2 diabetes.

Body composition — your ratio of different kinds of fat, skeletal muscle, and bone — is also an excellent marker of insulin resistance and metabolic health, so body scans are often used to help determine insulin resistance. Body comp is often measured with medical-grade 'DXA' scans, which use X-Rays, or commercially available devices at home.

These home devices can be the best smart scales or even some smartwatches, such as the Samsung Galaxy Watch Ultra 2. Without medical DXA scanners, they usually use a process called 'bioelectrical impedence analysis', or BIA. A low-level electric current is sent through the user's body, moving through fat, muscle, and bone at different speeds. The speed of the electrical current helps the device estimate the percentages of each material in the body: a useful thing to have on a health-monitoring device like a smartwatch.

(Image credit: Google)

However, Google is busy researching an even more accurate version of a body scan without using any specialist equipment or BIA kit — in fact, it's just a smartphone camera and an onboard AI model. This in-development feature is called PhotoScan, appropriately.

According to its research blog, Google has been developing PhotoScan by matching smartphone photographs of subjects with medical-grade DXA scans, along with other subject information, then running it all through an AI model. The AI model eventually achieved 'strong DXA agreement' when estimating body fat percentage and other metrics, leading to 'near-DXA accuracy for predicting insulin resistance'.

The researchers conclude that 'clinical DXA imaging delivers the most accurate body composition but lacks scalability, whereas wearable BIA sensors offer convenience but are limited to basic body fat percentage.

'Our PhotoScan approach offers a promising middle ground, estimating granular body composition from standard smartphone imagery with near-DXA accuracy.' Sorry, Samsung — it looks like we'll be pivoting away from BIA sensors and towards smartphone cameras for body fat percentage calculation in future.

While the technology doesn't seem to be commercially available yet, it does 'demonstrate the feasibility of smartphone-based body composition estimation,' and I'd be surprised not to see it make its way to the best Pixel phones eventually.

• Leaker Billbil-kun claims the Nintendo Switch 2 version of Diablo 4 will launch on September 15
• The "Age of Hatred Collection" will reportedly include the base game and the Vessel of Hatred and Lord of Hatred expansions
• The physical edition will have a code in the box

The long-rumored Nintendo Switch 2 port of Diablo 4 could reportedly launch very soon after BlizzCon 2026.

That's according to reliable leaker 'billbil-kun' of French outlet Dealabs (via Resetera), who originally reported on the Switch 2 version of Diablo 4 earlier this month.

Now, they claim that Diablo 4 will arrive on Switch 2 on September 15, nearly three years after its release on PS5, PS4, Xbox Series X, Xbox Series S, Xbox One, and PC, two days after BlizzCon 2026, which takes place between September 12 and September 13.

The edition will be called the "Age of Hatred Collection," and will include the base game and its two expansions, Vessel of Hatred and Lord of Hatred, and will cost $69.99 / £62.99 / €69.99.

Pre-orders are expected to begin on September 12, coinciding with the opening of BlizzCon 2026, and it appears that the physical version won't have a cartridge, but will offer a code in a box instead.

Billbil-kun couldn't determine when the game will be announced, but suggested that an official reveal could be made during Gamescom 2026 next week, where Xbox has confirmed its presence, but it's currently unclear. The announcement could just as well arrive at BlizzCon, too.

Diablo 4 launched in 2023 and is regarded as one of the best isometric, open-world role-playing games (RPGs). In TechRadar Gaming's five-star review, freelance contributor Cat Bussell praised the game's "darkly compelling fantasy world," "fluid combat that scales well in single and multiplayer," and "endlessly customizable characters."

Jack Reacher is the action hero that I never expected to fall in love with — but if you've been a long-time fan, you probably knew that would be the case long before I did.

As of writing, Reacher season 4 is dominating the Prime Video charts, and for good reason. Without giving anything away, the action has ramped up in a way that the Amazon TV adaptation has never seen before, and with a fifth season having already been renewed, things aren't about to slow down any time soon.

But things didn't start with a TV show a few years ago, did they? The character of Jack Reacher has been around for much longer, and has quite literally taken on a variety of different faces.

Here's where the challenge comes in. Whether you're a Reacher newbie or seasoned veteran (military pun intentional), just how much do you know about the man behind the action legend? Take the quiz below to find out:

As a side note, you'll need to have seen Reacher seasons 1-3 to get as many questions as possible correct, alongside having at least a baseline understanding of season 4.

While there's some crossover with the book series, questions are primarily based on the Prime Video series — I am a streaming writer, after all.

Even so, I haven't gone easy on you... I'd bet any imaginary money that you won't score about 70%. Prove me wrong, dear reader, prove me wrong.

Air fryers can be useful in all kinds of kitchen setups, but they're especially worth considering if you're moving into student accommodation. An air fryer can almost completely replace an oven and stove, performing of a whole spectrum of frying and baking jobs with ease. They're perfect for cooking small portions quickly, without wasting excess energy too.

Our general best air fryer guide runs down more options in more detail, but below I've rounded up my top recommendations for students in particular. I've focused on affordable options, and included a number of small-footprint fryers for those who find themselves fighting for counter space in a pokey student kitchen. I've also featured the Ninja Crispi, which uses the same glass containers for cooking, serving and storing, making it an extremely low-effort (and low-washing up!) option. Scroll down for my top student air fryer picks, or hop to the bottom for my buying tips to help you choose.

Student suppers

Ninja Crispi Glass Air Fryer

Read our full review

Pros

• Portable
• Four cooking modes
• Glass cooking dishes in multiple sizes
• Cook, serve, and store food in the glass dishes
• Easy to clean

Cons

• Baking mode isn’t great
• Not particularly attractive on the counter

Instant Pot 4QT Vortex Plus 6-In-1 Mini Air Fryer

Instant Pot 4QT Vortex 6-In-1 Mini Air Fryer with window

BELLA Fits-Anywhere Slim 4QT Air Fryer

Chefman Crispinator 6-in-1 Air Fryer, 8QT

Cosori Lite 2.1 QT Small Air Fryer

Ninja 4QT 4-in-1 Air Fryer

Cosori Turboblaze Air Fryer, 9-In-1, 6QT

Ninja Air Fryer Pro, 5QT, 4-in-1

How to choose

At this point, the air fryer market is saturated with options. If you're not quite sure how to choose the right one for you, here are some things to focus on to narrow down your options:

1. Available counter space

Consider the air fryer's shape and footprint — you need something that's not going to take up all your counter space. Remember, too, that you should be leaving some breathing space to the sides and rear of the appliance. If you're short on space, look for a tall design with edges that go straight up (rather than bowing out).

2. Cooking capacity

For one or two people, 4QTs will be just fine. However, if you want to cook for more people, you'll want something bigger. For more complex meals, consider a dual-zone air fryer with two drawers that can be set to different temperatures.

3. Brand reliability

There are some really cheap, no-brand air fryers on Amazon, but if you want an appliance that'll last then I'd stick to a brand that's established in the kitchen appliance arena. Ninja is often touted as the leader in this market, and it makes some excellent air fryers, but we've also been impressed by options from the likes of Instant Pot and Cosori.

4. Function options

A lot of air fryers will be advertised as being able to deliver multiple functions in one, and will offer functions like baking, roasting and crisping alongside air frying. I feel these are an unnecessary gimmick — with my own air fryer, I never use anything but the air fry function (and if I'm quite honest, I struggle to see how the process of baking will be different to that of roasting; it's all just heat being circulated in the cavity at a specific temperature). I wouldn't factor it too heavily into your choice-making.

• Alibaba Cloud's Lingjun Zhenwu M890 supernode instance is now on commercial sale
• It offers rentable 64-card computing units built entirely on in-house silicon
• It marks the first time Alibaba has directly sold supernode compute through the public cloud, however availability is limited to the Ulanqab region in Inner Mongolia

Alibaba Cloud has begun commercial sales of its Lingjun Zhenwu M890 supernode instance, letting customers rent a 64-card, high-speed interconnected computing unit without building their own data center.

The launch makes it the only Chinese-origin supernode that users can effectively use online without entering a purchase agreement or using a preconfigured AI model instance, an impressive enterprise-centric addition.

There is a caveat for enthusiasts looking to test it outside China: it is currently limited to users in North China's Inner Mongolia Autonomous Region only.

What is Alibaba offering to enterprise consumers?

A supernode is a set of accelerators wired together tightly enough that software treats them as one very large chip rather than a network of separate ones. This approach is often necessary; a model too big to fit on a single card would otherwise have to shovel data across comparatively slow networking, and that overhead can determine whether trillion-parameter models are practical to serve.

One of the best examples is Nvidia's GB200 NVL72, which leverages 72 of its Blackwell GPUs in a single configuration, or AMD's 72-GPU Helios configuration, which offers its own Instinct MI455X GPUs with more memory per GPU in tow.

Alibaba's offering is not faster than either example, and both are among the fastest options available to consumers worldwide, but it offers a mix of advantages to domestic Chinese consumers that would not otherwise be in play: Chinese Silicon, a Chinese cloud, and lower latency.

A domestic supernode built around a mixture of state restrictions and sanctions

As Beijing restricts Chinese cloud providers from accessing the latest Nvidia and AMD chips they can buy in a tit-for-tat move for the White House's increasing heavy-handedness, the best options for enterprise customers and AI startups have become increasingly domestic, and Alibaba Cloud's homegrown silicon is a key example of how that policy has shaped its offerings.

The underlying Zhenwu M890 processor debuted at the Alibaba Cloud Summit in May 2026, where the company also disclosed a detailed chip roadmap through 2028. comes from T-Head, the group's chip design unit, and carries 144GB of HBM memory with inter-chip bandwidth up to 800 GB/s. The interconnect fabric uses an in-house ICNSwitch 1.0 chip, which lifts the scale-up domain from 16 cards on the previous-generation Zhenwu 810E to 64, with bandwidth up from 700 GB/s.

The company says that the configuration is, on paper, capable of handling a Mixture of Experts (MoE) AI configuration of up to 10 trillion parameters, while emphasizing that this is the first supernode-form architecture in China to successfully run large language models exceeding 2 trillion parameters; Alibaba says the current instance is already powering commercial services for Kimi K3 and Qwen3.8-Max.

Alibaba's roadmap continues past the M890. T-Head has said the Zhenwu V900 arrives in the third quarter of 2027 with three times the M890's performance, 216GB of memory, and 1,200 GB/s of inter-chip bandwidth, followed by the Zhenwu J900 in the third quarter of 2028.

There is no public pricing for the instance, and information on its power consumption is nonexistent; benchmarks comparing it to mainstream options are scant at best. Despite the limited information and geographic limits, the move underscores both China's growing self-reliance in the AI industry and its ambitions, which could see the supernode offered to a wider range of consumers, both domestic and international, in the near future.

Google hasn’t confirmed the brand-new, unreleased wearable, but it’s fun to speculate anyway.

• Nvidia's open source NeMo Switchyard 'smartly' routes each agent request to the cheapest model that can handle it
• The approach allows it to claim a 74% cost cut against a frontier-only model in return for a 6% reduction in accuracy
• NeMo Switchyard joins an increasingly well-established territory that already sees players such as RouteLLM, LiteLLM, and OpenRouter, in addition to in-house efforts at OpenAI and AT&T to save costs

Nvidia has released NeMo Switchyard, an open source model router that sits between an application and a pool of language models and decides, per request or per turn, which model should handle a particular task.

The aim is to maximize efficiency by picking the right model for the right task: pushing frontier-level models to do simple tasks that much smaller or cheaper models could handle is not only inefficient, but also costly for enterprise customers.

This comes at a time when enterprise customers are already grappling with rising costs that can quickly spiral out of control as they increasingly adopt AI.

An open-source addition to a growing chorus of AI routers

The problem Nvidia is trying to solve isn't new, and industry giants are already attracting a lot of attention. The emerging field itself is potentially lucrative, even if it focuses on cutting costs, and major players are looking to get in on the action; Stripe's upcoming $7+ billion acquisition of OpenRouter, underscores this.

Switchyard is no different from its peers; it is essentially a proxy that uses a routing algorithm to decide which AI model to send queries to, while delivering an answer in the format that the calling application or user requires.

It accepts OpenAI, Anthropic, and Responses API requests, translates between them, and documents the selected model, decision rationale, token usage, and latency for each call, enabling transparency and letting it tweak or tune its approach over time.

This makes sense compared with a one-size-fits-all approach that would otherwise be prevalent in a world with limited frontier-level AI compute, as LLMs continue to grow larger and more demanding. One example is what Nvidia points to: Nemotron Parse, a one-billion-parameter model primarily designed to extract structure from PDFs.

The question remains, however, whether such an approach is always feasible. Leveraging such a routing tool comes at a cost; LangChain's test of Nemotron 3.5 Lightning found that its judge model, which determines whether the agent is still on track after each turn, consumed up to 21.2% of total cost, second only to what it spent on Claude's Opus 4.8.

Running a more optimized judge model could therefore yield greater savings or lower overhead, but it could also hinder cases where users need a more basic AI model, and the judge model increases costs by reading or checking output at every turn.

A secondary concern would be variance: while costs are noticeably lower than running just Opus 4.8 in testing, it does swing from anywhere between $2.16 and $3.61, a mammoth 67% movement essentially guided by when it escalated queries to a bigger or smarter model. This tradeoff makes cost hard to predict: while routing lowers average spend, it widens the distribution around it.

LangChain also says users should not use Switchyard if their workloads are short or latency-sensitive; it estimates it adds about 700ms of latency because the judge model has to read all output.

Nvidia's take here is simple: it is pushing a packaged, open-source solution that offers a broad set of integrations in what is essentially a relatively fragmented space. Routing work to smaller open-weight models pushes inference toward hardware enterprises own, which Nvidia sells.

At the same time, cheaper inference per task has historically meant more inference, not less, which is the outcome a company selling accelerators would choose to woo enterprises, many of which are still on the fence when it comes to directly investing in the hardware that powers it all.

Polaroid turned Pikachu, Sylveon and the Poke Ball into instant cameras because slapping Pokemon on things remains an excellent business model.