Connect with us
DAPA Banner
DAPA Coin
DAPA
COIN PAYMENT ASSET
PRIVACY · BLOCKDAG · HOMOMORPHIC ENCRYPTION · RUST
ElGamal Encrypted MINE DAPA
🚫 GENESIS SOLD OUT
DAPAPAY COMING

Tech

Stats watchdog prescribes stronger caveats for NHS Palantir claims

Published

on

PUBLIC SECTOR

Health service told to ensure ministers don’t mistake before-and-after comparisons for cause and effect

The UK’s statistical regulator has told NHS England to ensure ministers are briefed accurately following a row over data used to promote the purported benefits of a Palantir-run platform.

The Office for Statistics Regulation (OSR) said the health quango had accepted that it needed to communicate more clearly about data purporting to measure the benefits of the NHS Federated Data Platform (FDP), which is based on the US spy-tech firm’s technology under a £330 million contract.

Advertisement

“We expect NHS England to ensure that any figures cited by ministers or in official briefings about the FDP and its effectiveness are communicated in a clear, accurate manner,” the OSR said. 

Campaign group Foxglove called on health officials and ministers to correct the record with Parliament and the media on data used to illustrate the FDP’s success.

Earlier this month, the OSR, part of the UK Statistics Authority, said it was assessing how NHS England used data to make claims about the FDP’s performance, which sometimes made its way into ministerial statements.

The move followed a Freedom of Information request from Foxglove, which found that nearly a third of NHS trusts using Palantir’s health data platform were performing fewer patient procedures than before it went live.

Advertisement

Foxglove also said Chelsea and Westminster Hospital NHS Foundation Trust accounted for 84 percent of the reported fall in outpatient waiting lists, while only 16 trusts were using the tool provided by Palantir.

The results sit uneasily alongside NHS England’s claim that, as of June, 139 trusts used the FDP, with 137 reporting benefits. Another commonly quoted data item found that an Inpatient Care Coordination Solution (CCS) tool based on the FDP had resulted in 111,589 additional patients undergoing procedures in operating theaters.

Writing to James Murray, Secretary of State for Health and Social Care, Foxglove noted that NHS England had said in documentation that it was inappropriate to “draw conclusions about cause and effect as other variables have not been controlled for” in its published FDP data.

Foxglove said: “The claim that Palantir’s FDP has delivered a large number of additional operations has been widely repeated.”

Advertisement

At the time, the OSR said it was assessing the information around NHS England’s claims about the FDP’s performance.

Publishing its response yesterday, the regulator said it had engaged NHS England statisticians and conducted its own desk research based on information provided by the NHS and others who raised concerns about data used to promote the FDP’s benefits.

It noted that on June 6, NHS England added a sentence to published data, which read: “We cannot therefore draw conclusions about cause and effect as other variables have not been controlled for.” It also updated the methods page.

“NHS England has told OSR that this was motivated, in part, by NHS England having received numerous Freedom of Information requests to provide more detail on how data provided by FDP was analysed,” the OSR said.

Advertisement

“While the addition of this sentence is welcome, we understand why many users took this to represent a significant change in how NHS England was communicating the benefits of FDP data in relation to NHS performance.”

The statement welcomed NHS England’s decision to acknowledge that there is “more that can be done to strengthen messaging around the FDP benefits data and its onward interpretation by users.”

NHS England agreed to add further caveats to the published data and ensure public communications clearly identify before-and-after comparisons and explain that they do not establish causality.

It has also commissioned Imperial College London to evaluate the FDP, looking at products including OPTICA “where the issue around causality is most evident.” OPTICA is designed to streamline the patient discharge process. It was based on Palantir’s technology built under a £60 million series of contracts awarded without competition and then transferred to the FDP. NHS England said it would publish the results.

Advertisement

The OSR said it wanted to be kept informed of progress against a series of other commitments by NHS England’s Head of Profession for Statistics.

An NHS England spokesperson said: “We welcome this review from the Office for Statistics Regulation into the presentation and communication of information about the Federated Data Platform, which found the updates we made to our website have helped to make information more explicitly clear. We are implementing the recommendations in full to provide further clarity about the patient benefits being achieved through the use of the technology.”

Donald Campbell, advocacy director at Foxglove, said: “This damning report makes it clearer than ever that the Government and Palantir must withdraw these dodgy marketing claims and correct the record.

“Even the head of NHS England has admitted personally raising doubts about the claims made on Palantir’s benefits to the NHS. This intervention by the stats watchdog should be the nail in the coffin for the fairy stories we’ve heard about the extra operations supposedly delivered by Palantir’s platform.

Advertisement

“With a new government, there’s a chance to draw a line under the previous administration’s shameful practice of running marketing for this Trump-aligned tech giant. The new health secretary must withdraw these figures, correct the record with Parliament and the media, and ensure that the debate about whether Palantir should have a place in our NHS is conducted based on reliable, verifiable evidence – not spin.” ®

Source link

Continue Reading
Click to comment

You must be logged in to post a comment Login

Leave a Reply

Tech

Hide My Email, price rises, Apple Upgrade, and mass iPhone production starts

Published

on

The Hide My Email failure is real but nowhere near as significant to users as it’s being portrayed, Apple is hiking so many prices yet looking at ways to make that palatable, plus Foxconn has begun its annual recruitment drive as it begins producing millions of new iPhones, all on the AppleInsider Podcast.

It’s been painted as a big security failing by Apple, and it definitely isn’t good. But even if someone does manage to circumvent your use of Hide My Email, the worst they can do is send you a spam message.

There’s more to it, of course, and if there weren’t then you wouldn’t benefit from listening to Wesley Hilliard explaining it on the podcast. But while it’s good to know the background, not to mention interesting, you can be reassured too.

Which you might also be by how Foxconn is ramping up its recruitment right on schedule. Even if you’re not looking for a job assembling iPhones, it means that the iPhone 18 Pro is on track.

Advertisement

It would be, of course, and the only real questions are whether it can possibly match the success of the iPhone 17 Pro range, and whether there will be an iPhone Fold.

Although of course, there is also the issue of just how much any of these new iPhones will cost. But a strong new rumor claims that Apple will shortly launch a new Apple Upgrade program, specifically to make it possible to buy with a lower upfront cost.

If Apple gives us this leasing program, it also may take away something. To protect it from people just signing up and walking away with costly iPhones, Apple is believed to be readying a way to restrict the use of its devices if a payment is missed.

BONUS: Subscribe via Patreon or Apple Podcasts to hear AppleInsider+, the extended edition. This time, speaking of payments, the discussion is on technology and money, specifically how we need to manage our finances and how options like Apple Card can usually help.

Advertisement

More AppleInsider podcasts

Tune in to our Smart Home Insider podcast covering the latest news, products, apps, and everything HomeKit related. Subscribe in Apple Podcasts, Overcast, or just search for HomeKit Insider wherever you get your podcasts.

Podcast artwork from Basic Apple Guy. Download the free wallpaper pack here.

Those interested in sponsoring the show can reach out to us at: [email protected].

Subscribe to AppleInsider on:

Keep up with everything Apple in the weekly AppleInsider Podcast. Just say, “Hey, Siri,” to your HomePod mini and ask for these podcasts, and our latest HomeKit Insider episode too. If you want an ad-free main AppleInsider Podcast experience, you can support the AppleInsider podcast by subscribing for $5 per month through Apple’s Podcasts app, or via Patreon if you prefer any other podcast player.

Advertisement

Source link

Advertisement
Continue Reading

Tech

Slopsquatting, Phantom Domains, and HalluSquatting Are the Same AI Attack

Published

on

Cyber locks

Three attacks, three names, and one identical flaw: AI coding agents treat a hallucinated identifier as a verified command.

By Shane Warden, Principal Architect, ActiveState

Ask an AI coding agent to fetch a tool. Occasionally, it returns a name that sounds right, but does not exist. Developers used to ignore this mistake, assuming a compiler or test would stop it. Unfortunately, that assumption is dangerous and wrong.

Effectively now we’re giving root access to language models that sometimes guess words, and attackers know how to use that against us.

Advertisement

The attack goes like this: an attacker can calculate URLs, software library names, and other output an LLM will produce and subsequently access somehow. The attacker grabs the name, then sets a trap and waits.They do not need to steal passwords. They do not need to send phishing emails.

They do not need a human to click a link. They need someone, somewhere to give an automated process permission to fetch something malicious.

Researchers at Tel Aviv University, Technion, and Intuit published a paper on July 8, 2026. The team, led by Aya Spira in Ben Nassi’s group, proved that these fake names are predictable.

They tested multiple prompts across Cursor, Windsurf, GitHub Copilot, Cline, Gemini CLI, and OpenClaw. Models hallucinated identical names up to 85% of the time for repository requests. They hallucinated identical names 100% of the time for skill installs.

Advertisement

“Slopsquatting exploited a fake package name. Phantom squatting exploited a fake domain. HalluSquatting exploits a fake repository or skill,” Warden says. “In every case, the agent trusts a name nobody verified.”

This Time, the Payoff Is a Botnet

That comparison undersells what is new here. Slopsquatting and phantom squatting get an attacker’s code onto one machine at a time. HalluSquatting scales further for attackers, because the agent does the delivery work a botnet operator used to need real machine compromise for.

There are no stolen passwords, no worms crawling from device to device, and no single operating system to target. Any machine running an exposed agent can be a target.

The researchers built this attack specifically to demonstrate that it repeats at scale, which is why they treat their own numbers as a floor rather than a ceiling: “Attacks always get better,” they write. “They never get worse.”

Advertisement

Cordyceps passed every check because no single workflow file was wrong, it was the composition that was exploitable.

See how to close that gap by governing what enters your build at the source, not just what passes the scan.

Close the Gap

The Third Time Is a Pattern, Not a Coincidence

Security teams have seen this failure three times in six months.

The late-binding attack pattern

January 2026 (Slopsquatting): Charlie Eriksen at Aikido Security caught AI agents trying to install a fake npm package named react-codeshift across 237 projects. He registered the name to stop attackers from taking it.

Advertisement

June 2026 (Phantom Squatting): Unit 42 at Palo Alto Networks found 250,000 domains that language models hallucinate. Anyone could register those domains.

July 2026 (HalluSquatting): Spira’s research team showed how attackers can take over AI agents by claiming predictable names in advance.

Every attack relies on the same core defect. The system trusts a name that a model generated without checking if anyone verified and validated it.

Design Principle Failure: Late Binding

This problem stems from two connected issues.

Advertisement

First, language models generate their outputs based on probabilities. Given the right inputs, they’ll produce predictable outputs (not deterministic, of course, but the paper shows that they’re predictable enough to produce attack vectors). This reminds me of rainbow tables, where you can pre-generate the output of password hashes.

You don’t have to know someone’s password if you can get the (insecure) hash and you know an input which creates that hash.

Second, developers build and run code and pipelines that execute commands based on the generated output of these language models. Unless you’re very careful, your agents will run code or fetch data before they take the opportunity to vet and verify the source of that code and data.

In software architecture, this kind of late binding offers flexibility. In security architecture, trusting an unverified external data source creates a massive vulnerability. The danger lives in the gap between text generation and code execution. What was safe and true may have changed since you last checked yesterday (or an hour ago).

Advertisement

Developers are rushing to ship products faster. They treat build infrastructure like disposable tooling, even while granting these tools expansive permissions to download, change, and deploy things. They have automated typosquatting and dependency confusion.

When agents now have permission to fetch and execute code without sufficient human review, your risks of attacks grow.

The Risk Extends Down the Dependency Tree

The risk extends beyond the top-level packages. An agent might select a real package with a real name. Modern security tools may check that top-level package. Do you know they rarely inspect the transitive dependencies three or four layers down the tree?

Developers cannot inspect those dependencies simply by reading the top-level source code or saying “That looks correct” and hitting Enter. Worse, if a dependency is compromised but the previous versions were fine, the rules you had in place yesterday may not protect you today.

Advertisement

If an attacker can compromise a deep dependency or anything it relies on, an automated pipeline can bring that compromise into your systems.

Traditional Security Backstops Fall Flat

The researchers note that their findings represent a minimum risk level. These attacks will become faster and more accurate in their targeting. Existing security tools fail against these attack patterns.

In June 2026, Trail of Bits bypassed agent skill store scanners in less than an hour. Scanners examine stated claims rather than hidden payloads.

SSL certificates and DNSSEC fail to stop this threat. An attacker who registers a fake domain can easily get a free Let’s Encrypt certificate. The certificate proves who owns the domain, but it cannot prove that the user intended to connect to it or that the domain is safe.

Advertisement

DNSSEC prevents other people from taking over a domain, but what if the domain were registered yesterday because an attacker predicted the latest model would send people there?

Fixing the Intake Pipeline

To secure your systems, you must ensure that none of your pipelines ever execute unvetted code or data. That vetting and verification has to happen automatically, at the speed of AI. Human review cannot keep up with automated AI tools.

Engineering teams must address this problem directly. They can spend significant time building internal verification pipelines, or they can adopt an existing governance solution.

Teams that patch individual tools will spend years chasing new variations of this exploit. Teams that fix the underlying design flaw will stop the attack before the model ever runs an untrustworthy command.

Advertisement

Flowchart

Organizations must resolve open source dependencies through something like ActiveState’s Curated Catalog, a private, policy-governed repository of vetted components. The catalog verifies the package before the agent downloads it. A package which fails this vetting is completely invisible to the agent, causing a failure before any bad code can enter your systems. 

This is a different defense than scanning. Trail of Bits broke scanners because scanners inspect an unknown upload at the moment of fetch, exactly when an attacker has optimized the payload to slip past.

The Curated Catalog removes that moment entirely: it only ever serves components that were vetted and verified before any agent asked for them, so there is no unknown upload left to bypass. This single step turns a statistical guess into a trusted resource or a deliberate failure which needs human investigation.

If you want to see how many of your team’s current fetches would pass unchecked today, request an OSS Risk Assessment from ActiveState and find out before an attacker does.

 

FAQ

What is HalluSquatting?

HalluSquatting is an attack where researchers pre-compute the fake repository, package, or skill names that AI coding agents predictably invent, register those names first, and load them with malicious instructions before a real user’s agent goes looking for them.

Advertisement

How is this different from slopsquatting or phantom squatting?

All three exploit the same flaw, an agent trusting a name nobody verified, but target different resources. Slopsquatting targets npm package names, phantom squatting targets web domains, and HalluSquatting targets repositories and agent skills, then executes the payload directly through the agent’s own tool-use permissions.

Can existing security scanners catch this?

Not reliably. Trail of Bits bypassed every public skill-store scanner they tested in under an hour, because scanners inspect an unknown upload’s stated content rather than its hidden payload. A static, post-hoc scan is racing an attacker who built the payload specifically to defeat that scan.

What should engineering teams do this week?

Turn on pre-fetch verification wherever it exists; most agent frameworks ship with it off by default. Route open source dependency resolution through a governed, pre-vetted catalog instead of letting agents fetch directly from public registries.

Does this affect every AI coding agent equally?

The researchers found hallucinated names were consistent across tools built on different underlying models, including Cursor, Windsurf, GitHub Copilot, Cline, Gemini CLI, and OpenClaw. This is a pattern in how agents are built and permissioned, not a flaw isolated to one vendor.

Advertisement

Sponsored and written by ActiveState.

Source link

Advertisement
Continue Reading

Tech

SpaceX to try its luck again with Starship Flight 13 after engines and weather say no

Published

on

SCIENCE

Replacement Raptors head for liftoff without a static fire test

SpaceX is gearing up for another attempt to launch Starship following a last-second abort on July 16 and a scrub due to weather yesterday.

The 90-minute launch window for Starship’s 13th flight test will open at 1745 CT on Friday, July 24. The launch was postponed by 24 hours due to weather, according to SpaceX.

Advertisement

Elon Musk’s rocketeers explained: “A key objective for the flight test is to get clear imagery from the ground of Starship’s heatshield as it flies at a higher dynamic pressure during ascent, which won’t be possible with today’s weather conditions. Visibility is forecast to be ideal for a Friday attempt.”

The first attempt last week was aborted in the final seconds after some of the booster’s Raptor engines failed to start. At the time, Musk said “two Raptors will be removed and replaced.”

Interestingly, the company did not perform a static fire test of the booster with the new engines on July 22, opting for what appeared to be little more than a leak check. The Register asked SpaceX what testing it performed, but it has yet to respond. In a post on X, the company only stated “additional preflight testing on Super Heavy complete.” There is speculation that last-second shutdowns, whether during a static fire or launch attempt, may not be kind to Raptor engines.

The Starship project is under greater scrutiny than ever. NASA requires it to have orbital capability for Artemis III in 2027, and a launch cadence high enough to support a lunar landing mission in 2028. SpaceX is also a publicly traded company and, while its current stock price has already fallen below its IPO, it could do without further rattling investor nerves with a failure.

Advertisement

Should all go well, the entire mission should last just over an hour. After launching Starship, the Super Heavy booster is expected to make a controlled splashdown in the Gulf of Mexico. Starship will use its Raptor engines to enter a suborbital trajectory before coming down in the Indian Ocean.

Before re-entry, SpaceX plans to deploy 20 Starlink V3 satellites, which will follow the vehicle on its suborbital trajectory, and attempt a relight of a Raptor engine. The latter is a critical step on the path to Starship going orbital. ®

Source link

Advertisement
Continue Reading

Tech

SoftBank is reportedly weighing a deal for Swiss robotics startup Gravis

Published

on

Gravis Robotics, an ETH Zurich spinout, fits autonomy kits to excavators. A tie-up would extend Masayoshi Son’s robotics buying spree from the factory floor to the building site.


SoftBank is weighing a deal for Gravis Robotics, a Swiss startup that fits excavators and diggers with the sensors and software to run themselves, according to Bloomberg.

The report did not spell out the shape of any transaction, and the size, structure, and valuation were not disclosed. The discussions appear to be at an early stage, and there is no guarantee they lead anywhere.

What they do signal is where SoftBank is looking next. Masayoshi Son’s group has spent the past year assembling a robotics portfolio at speed, from an $800mn round it is reportedly circling for Agile Robots to smaller bets on the kind of autonomy kits that bolt onto existing machinery.

Advertisement

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

Gravis Robotics spun out of ETH Zurich in 2022, from the Swiss university’s Robotic Systems Lab. Rather than build a robot from scratch, it retrofits standard heavy equipment, attaching a kit that fuses LiDAR, cameras, GNSS positioning, and hydraulic sensors so an ordinary excavator can trench, grade soil, and manage stockpiles on its own.

The company says the approach can lift site output by roughly 30%.

Advertisement

The kit has a name, or two. The hardware, Gravis calls Rack; the operator controls it through a tablet interface named Slate that switches between autonomous and manual modes.

The pitch is that a machine can be taught to feel the soil through its hydraulics rather than follow a fixed programme.

It is run by chief executive Ryan Luke Johns, an architect turned roboticist, alongside chief technology officer Dominic Jud and co-founder Marco Hutter, an ETH robotics professor.

Johns and Jud hold a Guinness World Record together for the largest robot-built dry-stone wall, which tells you something about the company’s temperament.

Advertisement

In November 2025 the startup raised $23mn in a round co-led by IQ Capital and Zacua Ventures, with Pear VC, Sunna Ventures, and cement group Holcim among the backers.

By its own account, Gravis now has machines working on active sites across seven countries. Its customer list, per the company’s own announcements, includes contractor Taylor Woodrow, which used the technology at Manchester Airport, plant-hire firm Flannery, and Holcim, which runs it in quarries. Those claims come from Gravis and have not been independently audited.

For SoftBank, construction is a logical extension of a much larger thesis. The group agreed last October to buy ABB’s robotics division for $5.4bn, and a four-bank syndicate only finished putting together the roughly $1.75bn loan behind that purchase this week.

It is also selling its remaining stake in Boston Dynamics to Hyundai, tidying up an older robotics bet even as it places newer ones.

Advertisement

The Gravis talks, if real, would join a run of similar moves. SoftBank led Agile Robots’ 2021 financing, the round that made the Munich firm Germany’s first robotics unicorn, and is now said to be anchoring a fresh $800mn raise there.

Earlier in 2025 it put $500mn into Skild AI, a startup building a general-purpose model to control robots. The common thread is machines that do physical work, and software good enough to run them.

Son’s pitch is that artificial intelligence has matured in software and the next frontier is giving it a body. SoftBank is reportedly preparing a US-based AI and robotics vehicle, provisionally called Roze, that it hopes to float at a $100bn valuation.

The wider market has moved with him, with global robotics investment more than doubling to $27.6bn in 2025, the year of record rounds for firms such as NEURA Robotics.

Advertisement

A digger that drives itself is a less photogenic proposition than a walking humanoid, but it may be a more immediately useful one. Construction faces a persistent shortage of skilled machine operators, and earthmoving is repetitive, dangerous, and expensive to staff.

Whether SoftBank ends up buying into Gravis, or simply kicking the tyres, the direction of travel is clear enough. The company that once bet on a chatty humanoid named Pepper now wants the machines that move actual earth.

Source link

Advertisement
Continue Reading

Tech

Narwal Flow 2 Review – Trusted Reviews

Published

on

Verdict

More suction, improved (and very good) obstacle avoidance and brilliant mopping with the track mop make the Narwal Flow 2 a great choice for those with hard floors. Improved suction power makes short work on carpet, but as the mop can only lift by 10mm, those with deeper pile carpets may prefer a model that can drop its pads or that close off their rollers.

  • Excellet mopping

  • Very powerful vacuuming

  • Moves around obstacles with ease

  • Can’t leave its mop behind for vacuum only mode

Key Features

  • Advertisement

    Track mop

    Wide mopping cloth polishes stains off easily.

  • Advertisement

    Powerful suction

    30,000Pa suction power makes short work of fine dust.

Introduction

With its unique mopping system, the original Narwal Flow proved to be the best robot at removing stains that I’d tested, but it wasn’t quite up there when it came to overall flexibility. With the Narwal Flow 2, the company is back to sort that.

Advertisement

There’s the same clever track mop system, but a lot more suction power and a redesigned docking station. Is it all enough to make the Flow 2 the best robot vacuum cleaner, or is more work required? Read my full review to find out.

Design and Features

  • Self-empty, self-clean dock
  • Track mop
  • Powerful obstacle avoidance

Advertisement

You can buy the Narwal Flow 2 in black or white, a choice that covers both the robot and its docking station. Aside from the way the robot looks, there’s no difference between the colour variants in terms of performance or features.

Narwal has completely redesigned its docking station for the Flow 2. The new one looks more like something that Roborock would ship, rather than continuing the more curved design of the original.

One change is that the new dock now has a front ramp that needs to be attached to the front, sticking out slightly. The old dock didn’t need this and contained the robot fully inside.

Advertisement

There’s also a status light bar on the new dock that glows to give a visual indicator of the robot and whether or not it’s ready to go or has any issues. That’s fine, but I’m actually a fan of fewer lights, and only having them when there’s an issue.

On top are the two water tanks. There’s a 4.5-litre tank for clean water, and a slightly smaller 4-litre tank for dirty water. Given that some water ends up on the floor after use, when the dirty tank needs emptying, the clean one will need filling.

Narwal Flow 2 water tanksNarwal Flow 2 water tanks
Image Credit (Trusted Reviews)

Advertisement

This docking station can use 100°C water for mop washing, plus it can use detergent, automatically mixing it. Narwal keeps the detergent holder underneath the water tanks. There’s a sample of the company’s own hard floor cleaner in the box, but the design means you can use your own cleaner if you prefer.

Narwal Flow 2 detergentNarwal Flow 2 detergent
Image Credit (Trusted Reviews)

Also under the tanks is the 2.5-litre disposable dust bag, which Narwal says can hold 120 days’ worth of dirt.

Narwal Flow 2 binNarwal Flow 2 bin
Image Credit (Trusted Reviews)

Pulling the main robot out, it’s quite similar to its predecessor in looks. With no LiDAR dome on top, the 95mm tall robot can slip under a lot of furniture without trouble.

Narwal Flow 2 top viewNarwal Flow 2 top view
Image Credit (Trusted Reviews)

Advertisement

At the front are the Dual RGB Cameras and Side Structured Light, which let the robot spot and avoid obstacles. Narwal says that this robot has unlimited obstacle detection, with Free FlowMove effectively allowing it to move around any issue and keep cleaning.

Advertisement
Narwal Flow 2 camerasNarwal Flow 2 cameras
Image Credit (Trusted Reviews)

There’s a small dust bin that can be removed to be cleaned out, which also contains the Flow 2’s filter.

Narwal Flow 2 robot binNarwal Flow 2 robot bin
Image Credit (Trusted Reviews)

On the underside is the FlowWash Mopping system. This shares some similarities with a roller, but the track mop system means there’s more surface contact with the floor. A cloth is dosed with clean water, and then pulled along the floor with 12N of pressure to remove dirt.

Narwal Flow 2 underneathNarwal Flow 2 underneath
Image Credit (Trusted Reviews)

Advertisement

In addition, the robot can use its cameras for stain detection, focusing on areas that need to be cleaned more.

When carpet is detected, the FlowWash system can be lifted 10mm off the ground to avoid soaking carpets. That works well enough on short pile carpet, but for more varied flooring you might prefer a robot that can either block off its roller or one that can drop its mopping pads off at the dock.

In addition, the Flow 2 can climb double thresholds of up to 4cm or single thresholds of up to 3cm, so it should be able to navigate most basic obstacles.

Control of the robot is via the Narwal app. It’s quick to get the robot connected to Wi-Fi, and then the initial mapping run. At the end of mapping, the app suggests the room layout, but I could merge or split to get the layout that I wanted.

Advertisement
Narwal Flow 2 app mapNarwal Flow 2 app map
Image Credit (Trusted Reviews)

Alongside the regular options, such as no-go zones, there are advanced options. Set the Room Character, for example, and you get to choose to say that a room is High-soil or Low-noise, which helps the robot adjust its cleaning strategy for that area.

Advertisement

Furniture can be added to the map, and then selected as a zone for quick cleaning, which is handy, say, for cleaning up after eating a meal. Otherwise, cleaning options including rooms of your choice or zones that you draw on the map.

Narwal Flow 2 appNarwal Flow 2 app
Image Credit (Trusted Reviews)

There’s a choice to vacuum and mop, vacuum then mop, vacuum, or mop. For each choice, you can use the Freo Min option, which cleans automatically based on the environment and detected dirt level, or you can override and choose your settings.

For vacuuming, there’s a choice of four power modes, and two coverage choices (depending on how thorough you want the robot to be), plus up to three passes. Mopping has the same basic coverage settings but three choices of mop humidity.

Narwal Flow 2 app cleaning optionsNarwal Flow 2 app cleaning options
Image Credit (Trusted Reviews)

I like that with the vacuum-then-mop setting that I could pick different numbers of passes for vacuuming (one or two) to mopping (up to three). This makes it much easier to customise how the clean works.

Advertisement

Advertisement

Performance

  • Excellent vacuuming power
  • Removes most stains fast
  • Brilliant navigation

I put the Narwal Flow 2 through its paces around the Trusted Reviews Home Technology Lab. Starting with the carpet test, I added a teaspoon of flour to the test carpet. With 30,000Pa of suction, this robot is one of the most powerful and it shows. After two passes, the carpet is back to its original state.

I had the same result on the hard floor with dust in the middle. This was all gone after vacuuming only.

My edge test proved tough in the vacuum-only test, with the side brushes not quite enough to tease out all of the dust. 

Advertisement

Things get a lot better when mopping is added, and the Flow 2 managed to run along my kitchen plinth mopping up that mess, leaving only a fine trace of dust behind. That’s as good as I’ve seen from any robot vacuum cleaner.

Advertisement

I then moved to the mopping tests, starting with the dried-on coffee stain. I had set the robot to three passes, but it only needed one to quickly wipe this up.

Similarly, the dried-on red wine stain was removed with just that single pass, leaving a streak-free finish.

Advertisement

My mud stain was first vacuumed, removing the loose debris, and then completely cleaned after two mopping passes.

Advertisement

My ketchup stain is the hardest to remove, and this did prove tricky, especially as the abnormally hot weather had really dried the stain out. After three passes, the stain had been hugely reduced, but was still visible.

Given how hard the stain had set, I gave it a quick spray with kitchen cleaner, left it for five minutes, and then let the Flow 2 go about its work. This time, it picked up everything: the mopping system really is impressive.

Narwal Flow 2 ketchup stain clean fullNarwal Flow 2 ketchup stain clean full
Image Credit (Trusted Reviews)

Advertisement

I put some human hair down on the carpet, and then let the Narwal Flow 2 collect the mess. Turning the robot over, there was one loose strand of hair, but nothing wrapped tightly around the roller.

Object detection and avoidance is very good, too. I didn’t have any issue with the robot running into fake pet mess, shoes or cables that I left out, neatly navigating around them while still cleaning.

Advertisement

Sound is good, too. I measured the robot at just 50.5dB, which makes it very quiet.

Battery life is as good as you’d expect, too. Even with three passes set, the Narwal Flow 2 could easily manage to clean the entire Lab with a single charge.

Advertisement

Should you buy it?

Advertisement

You want clever navigation and great mopping

The track mop quickly wipes up tough stains, while the new AI obstacle avoidance is a winner for cluttered homes.

Advertisement

You need more flexibility for carpets

If you’ve got deeper pile carpets, a robot that can drop its pads or close off its roller may be a better choice.

Advertisement

Final Thoughts

Its ability to move through cluttered areas with ease, combined with the powerful mopping performance, makes the Narwal Flow 2 a star on hard floors. Improved suction power makes for better carpet cleaning, too. The main downsides of this model are that it can’t drop its track mop or cover it when dealing with deeper-pile carpet, which may make it unsuitable for some homes, in which case an alternative from the list of best robot vacuum cleaners may be better.

How We Test

We test every robot vacuum cleaner we review thoroughly over an extended period of time. We use industry standard tests to compare features properly. We’ll always tell you what we find. We never, ever, accept money to review a product.

Find out more about how we test in our ethics policy.

Advertisement
  • Used as our main robot vacuum cleaner for the review period
  • We test for at least a week
  • Tested with real-world dirt in real-world situations for fair comparisons with other vacuum cleaners

FAQs

How does the Narwal Flow 2 deal with carpets?

When it detects carpet, the robot can lift its mop off the ground to pass; you can also set the app so that the robot avoids carpet.

Test Data

  Narwal Flow 2
Sound (high) 50.5 dB

Full Specs

  Narwal Flow 2 Review
Manufacturer
Size (Dimensions) 364 x 351 x 95 MM
Release Date 2026
First Reviewed Date 23/07/2026
Vacuum cleaner type Robot vacuum and mop
Bin capacity 2.5 litres
Modes Vacuum, vacuum and then mop, vacuum and mop, mop
Filters 1 (washable)
Run time 190 mins min
Charge time 3 hrs
Brushes 1x side brush, 2x brush bar
Mop Option Track mop
UK RRP TBC
USA RRP $1499.99

Advertisement

Advertisement

Source link

Continue Reading

Tech

Atari signs deal to turn Pong, Centipede, Asteroids, and more classic games into movies

Published

on

WTF?! As movie adaptations of video games have turned from mostly box-office bombs into commercial hits, it seems studios are now trawling the entire history of gaming for ideas. In what sounds like a particularly bewildering move, Atari has signed a deal with Universal Pictures to give 10 of the company’s classic games the movie treatment, including Pong and Centipede.

Atari’s retro games aren’t the first that come to mind when you think of titles that would be great on the big screen. The full list consists of Asteroids, Adventure, Berzerk, Breakout, Centipede, Crystal Castles, Millipede, Missile Command, Pong and Yars’ Revenge.

Deadline writes that Universal has already bought the first project based on an Atari IP, though it never revealed which one. It will be produced by Entertainment 360’s Guymon Casady from a screenplay by Matt Reilly and Carl Hampe.

“The best Atari games dropped you into a world and let your imagination do the rest,” Casady said. “Carl and Matt saw an opportunity to take that same spirit and build an original, large-scale adventure around it. From the moment we read the screenplay, we believed there was a great movie here.”

Advertisement

It’s certainly difficult to imagine how some of these games will translate into movies, but one expects there will be plenty of artistic interpretation. Perhaps Pong will be a Marty Supreme-style tale of becoming a table tennis champion. Missile Command could be a Cold War action thriller. Centipede might become a horror movie (nothing like The Human Centipede, presumably ). And Breakout could be a Shawshank Redemption-like prison-break tale.

The idea is to use the Atari IP to create large-scale action adventures, apparently. Hopefully, they’ll be better than that other “how could this be turned into a movie?” debacle, 2012’s Battleship.

Advertisement

Universal had previously secured the rights for an adaptation of Asteroids in 2019, but it still hasn’t been made into a movie. One assumes this would be an intergalactic adventure with aliens and a lot of threatening rock/metallic/icy space bodies.

Atari’s retro titles aren’t the only old games Hollywood is plundering for big-screen adaptations. Point-and-click classic Broken Sword, which first arrived in the mid-1990s, will also get a movie version, and it’s being made by the studio behind the live-action Sonic the Hedgehog movies.

Video game movies have come a long way since the days of Super Mario Bros in 1993 and the painfully awful Alone in the Dark. There are still a few stinkers, but when A Minecraft Movie made almost $1 billion, it’s easy to see why studios are desperate to make more.

Advertisement

Source link

Continue Reading

Tech

5 Biggest Samsung Launches from Galaxy Unpacked July 2026

Published

on

Samsung has expanded its Galaxy lineup with the launch of five new devices at its Galaxy Unpacked 2026 event in London. Three foldable handsets were unveiled, along with the Galaxy Watch 9 and Galaxy Watch Ultra 2, in addition to the Galaxy Z Fold 8 and Galaxy Z Fold 8 Ultra. The newest additions feature improved technology, slimmer form factors, Galaxy AI, and health monitoring capabilities. Preorders for the new products have already been opened in various markets, and they will be available starting from early August.

Samsung Galaxy Z Fold 8

galaxy Z fold8

The new Samsung Galaxy Z Fold 8 is built with people who prefer a larger foldable device in mind. For that reason, it comes with a passport-like design, which makes the device relatively shorter but gives a wider cover screen. This type of device enables easy completion of daily activities without unfolding it. The device has a 5.5-inch cover screen and a 7.6-inch foldable OLED screen. Both displays support an adaptive refresh rate from 1Hz to 120Hz.

The company uses a new Flex Titanium display structure to strengthen the folding screen. Samsung has redesigned the hinge for better performance and a less noticeable crease. It is powered by the Snapdragon 8 Elite Gen 5 chip. The device features up to 16 GB of RAM and 1 TB of internal memory. It includes a 50 MP primary sensor, a 50 MP ultra-wide lens, and a pair of 10 MP selfie shooters.

A 4,800mAh silicon-carbon battery keeps the phone running and supports 45W fast charging. The Galaxy Z Fold 8 ships with Android 17 and One UI 9. It also includes Galaxy AI tools such as Now Nudge, Gemini Intelligence, and Gemini Notebook. Buyers also get six months of Google AI Pro at no extra cost. The phone starts at $1,899.99 and is available in Lavender, Cream, Graphite, and an online-only Pistachio finish.

Samsung Galaxy Z Fold 8 Ultra

Galaxy Z Fold 8 Ultra

The latest addition to the line-up of the company’s top-selling foldable model is the Galaxy Z Fold 8 Ultra. The new smartphone replaces the Galaxy Z Fold 7 and offers a slimmer form factor. Its thickness measures 4.1 mm when unfolded, and 8.9 mm when folded. Samsung has also introduced a toughened Flex Titanium frame along with the improved hinge, which increases the strength of the display and minimizes the crease effect. The Fold 8 Ultra has a 6.5-inch outer screen and an 8-inch folding screen with a 120Hz adaptive refresh rate.

The Galaxy Z Fold 8 Ultra runs on the Snapdragon 8 Elite Gen 5 processor. It comes with up to 16GB of RAM and up to 1TB of storage. Samsung also upgrades the camera system. The phone includes a 200MP main camera, a 50MP ultra-wide camera, and a 10MP telephoto camera. It also has two 10MP selfie cameras, one on each display.

Advertisement

The Fold 8 Ultra includes a 5,000mAh silicon-carbon battery with 45W wired charging support. It runs Android 17 with One UI 9 and includes Galaxy AI features like Now Nudge, Gemini Intelligence, and Gemini Notebook. Samsung offers the phone in Violet Shadow, Cream, Graphite, and an online-only Green Shadow color. The device starts at $2,099.99, while preorder buyers can take advantage of Samsung credit and trade-in deals.

Samsung Galaxy Z Flip 8

Galaxy Z Flip8

The Galaxy Z Flip 8 is the latest folding clamshell from Samsung. The company retains its classic design, but now makes the device slimmer and even more advanced. The features of the new Galaxy Z Flip 8 include the 6.9-inch foldable display and the 4.1-inch FlexWindow cover display.

Galaxy Z Flip 8 comes with the Snapdragon 8 Elite Gen 5 processor in America. Overseas versions have the Samsung Exynos 2600 chipset. This model features 12GB RAM along with up to 512GB of internal memory. It has a camera system that comprises 50MP main, 12MP ultra-wide, and 10MP selfie cameras.

The phone runs Android 17 with One UI 9. Samsung includes Galaxy AI features that work directly on the FlexWindow display. Personalized Now Brief updates and Gemini Intelligence help users complete tasks more quickly. The Galaxy Z Flip 8 also packs a 4,300mAh battery with 25W wired charging. It is available in Pink, Cream, Graphite, and an online-only Mint finish. It starts at $1,199.99, and preorder buyers can take advantage of Samsung credit and trade-in deals.

Samsung Galaxy Watch Ultra 2

Galaxy Watch Ultra 2

The Galaxy Watch Ultra 2 is Samsung’s newest premium smartwatch. The watch features LTE connectivity as standard across all models. Samsung also redesigned the titanium body to make it thinner than before. The wearable includes a 47mm display with a peak brightness of 5,000 nits. It is powered by the Snapdragon Wear Elite platform and includes 2GB RAM and 64GB storage. Samsung equips the watch with a larger 800mAh battery. The company says it can deliver up to 60 hours of battery life with the Always-On Display enabled.

Samsung adds several new health and fitness features to the watch. These include Heart Health Score, Daily Cardio Load, Trail Run mode, and Nutrition Alert. The watch also provides hydration guidance during workouts. Samsung plans to release a Diving app later this year. The Galaxy Watch Ultra 2 has an IP69K rating and comes in Titanium Gray and Titanium Silver. It starts at $699.99, while preorder buyers can receive Samsung credit and trade-in savings.

Advertisement

Samsung Galaxy Watch 9

galaxy-watch-9-

Samsung has launched the Galaxy Watch 9 as its more affordable smartwatch. The watch comes in 40mm and 44mm sizes. Buyers can choose between Bluetooth and LTE models. The 40mm version is available in Cream and Graphite, and the 44mm model comes in Silver and Graphite. Samsung powers the watch with the Snapdragon Wear Elite processor. It includes 2GB RAM and 32GB storage. The display offers up to 3,000 nits of peak brightness. The 40mm model packs a 390mAh battery, while the 44mm version includes a 445mAh battery. Samsung claims up to 30 hours of battery life with the Always-On Display turned on.

The Galaxy Watch 9 introduces new health and fitness features. Users get Heart Health Score, Daily Cardio Load, Sleep Apnea detection, and Vitals tracking during sleep. The watch also warns users about unsafe environmental noise levels. Samsung gives the smartwatch MIL-STD-810H and IP68 ratings for durability. Pricing starts at $379.99, with LTE models priced $50 higher. Samsung also offers a 60-day Strava trial and a two-month iFIT membership for new buyers.

Source link

Continue Reading

Tech

Government Lawyers Say Trump Admin Can Use TikTok Again Because ‘Owned’ No Longer Means ‘Owned’

Published

on

from the words-mean-what-i-say-they-mean dept

We spent a few years pointing out the ridiculousness of the whole “TikTok ban” moral panic, and the fact that all of the “concerns” magically melted away after Trump became president and then effectively gifted a controlling stake to some of his friends should raise some pretty big questions. However, most people seem to have accepted the new arrangement without much fuss — even though ByteDance still retains a 19.9% stake in the company, and users at no point needed to switch to a brand new app, continuing instead to use the very app we were told was a security nightmare. All of which suggests the entire moral panic was absolute bullshit.

Either way, prior to the full “ban” that forced further ownership into the hands of Trump’s friends, there was a separate law from Senator Josh Hawley which simply banned TikTok on government devices. That law is still in effect. It’s pretty clear that it applies to “the social networking service TikTok or any successor application or service developed or provided by ByteDance Limited or an entity owned by ByteDance Limited.” It’s also clear that such an application is not allowed on any government devices, with exceptions only “for law enforcement activities, national security interests and activities, and security researchers.”

Now, a plain reading of the law would suggest that the current app is still banned. The law is still in place. ByteDance still owns a significant stake in the new “US joint venture” and the app is absolutely a “successor app” since users never needed to download a new app after the joint venture was established.

But, the Trump administration apparently would like to use TikTok on their devices. So, they’ve had the Office of Legal Counsel put out a decision claiming that, you know, ownership doesn’t really mean ownership and that the Trump administration can ignore the law and start using TikTok on their devices again. First things first, we discover that because Josh Hawley wrote a stupidly drafted law that directly called out “TikTok,” the OLC has to first tap dance around the fact that the law’s clearly named “TikTok” apparently doesn’t mean this TikTok, even though that’s exactly what the statute says:

Advertisement

Blackletter statutory-interpretation principles illuminate which particular “TikTok” Congress sought to prohibit. It is old wisdom that “a general phrase can be given a more focused meaning by the terms linked to it.” Fischer v. United States, 144 S. Ct. 2176, 2184 (2024). Namely, “the canon of noscitur a sociis teaches that a word is ‘given more precise content by the neighboring words with which it is associated.’” Id. at 2183 (quoting United States v. Williams, 553 U.S. 285, 294 (2008)). We apply this rule to “avoid ascribing to one word a meaning so broad that it is inconsistent with its accompanying words, thus giving ‘unintended breadth to the Acts of Congress.’” Gustafson v. Alloyd Co., 513 U.S. 561, 575 (1995) (quoting Jarecki v. G.D. Searle & Co., 367 U.S. 303, 307 (1961)). And precisely that kind of unexpected breadth would ensue here, were the Government Ban understood to apply to any future social networking platform based on its name alone.

[….]

We have considered the counterargument that, under the Dictionary Act, “words importing the singular include and apply to several . . . things,” 1 U.S.C. § 1—thus indicating that the Government Ban’s use of the phrase “the social networking service TikTok” could denote multiple unrelated variations or iterations of social media companies named TikTok. But the Dictionary Act itself provides that its general prescriptions do not apply when “context indicates otherwise,” id., and context does so in this case. “In context[,] the phrase ‘[the social networking service TikTok]’ should not be interpreted to mean literally ‘any [social networking service called TikTok],’ but must be understood against the background of what Congress was attempting to accomplish in enacting the [Government Ban].” Gustafson, 513 U.S. at 575 (cleaned up) (quoting Reves v. Ernst & Young, 494 U.S. 56, 63 (1990)). Here, the plain text of the Government Ban indicates Congress was attempting to address a particular national security threat posed by the presence on federal government devices of software “developed or provided by ByteDance Limited or an entity owned by ByteDance Limited.” Government Ban § 102(a)(1), 136 Stat. at 5258. TikTok USDS thus is covered by the ban only if it, like the version of TikTok operative when the ban was passed, falls into that category of software.

Call me pedantic, but if Congress didn’t want to ban an app “based on its name alone” maybe they shouldn’t have drafted and then passed a law that banned an app based on its name alone. And if Congress thinks that the new TikTok is somehow safer, they should repeal the original, poorly drafted law. Instead, the OLC has to start asking “what is ownership, really, other than a concept”?

For three reasons, we conclude that “ownership” in the context of the Government Ban is best understood as referring to a controlling stake, such that TikTok USDS falls outside the prohibition’s scope.

First, the “control” sense of the word “own” is most “consistent with the way that an appropriately informed speaker of the language would understand [that term’s] meaning” in the specific context of corporate structure. Van Buren v. United States, 141 S. Ct. 1648, 1657 (2021) (quotation marks omitted). The United States is home to “large numbers of firms with widely dispersed share ownership.” Henry Hansmann & Reinier Kraakman, The End of History for Corporate Law, 89 Geo. L.J. 439, 443 (2001). But it would be unusual for someone to say that a person or even an institutional investor “owns,” for example, Meta, simply because the investor holds some of its stock. Cf., e.g., Van Buren, 141 S. Ct. at 1657 (“In the computing context, ‘access’ references the act of entering a computer ‘system itself[.]’”). Instead, in the corporate context, we generally recognize Mark Zuckerburg as the “owner” of Meta because he retains control of the company through so-called “super-voting” shares. See Nathan Reiff, Top Facebook (Meta) Shareholders, Investopedia (Mar. 21, 2026), https://perma.cc/XQ6V-ZNTT; Gregory H. Shill, The Social Costs (and Benefits) of Dual-Class Stock, 75 Ala. L. Rev. 221, 224 & n.6 (2023).

Advertisement

So, hear me out, if Josh Hawley and Congress meant for the law to only apply if ByteDance “controlled” the company, then it could have (and arguably should have) written that into the law. But they did not. They said ownership. And that mattered because, technically before the “divestiture” and new US “joint venture” Western investors already owned about 60% of ByteDance, with employees and ByteDance’s founder holding most of the rest. The goal of the various laws to ban TikTok was to get ByteDance out of owning any of the company.

And that didn’t work. But we all have to pretend this “fixed” things, so the OLC just says “eh, because US entities now control it, we can ignore the law and pretend it said “control” rather than “own.”

Our textual interpretation is confirmed by the facts on the ground, which indicate that the TikTok USDS joint venture is wholly controlled by American interests as a functional as well as a formal matter—and thus exhibits none of the concerning security features that initially motivated the Government Ban. As our prior advice to you highlighted, if facts did not bear out that conclusion, then our understanding of “ownership” as used in the Government Ban could be called into question. But where, as here, the facts demonstrate that ByteDance Limited’s status as a minority shareholder in the joint venture has no impact on the exercise of control over the venture by United States investors, the inference runs the opposite way. Congress had no need to target minority ownership by ByteDance Limited in the Government Ban because that state of affairs is wholly compatible with the joint venture “operat[ing] [TikTok USDS] under defined safeguards that protect national security.”

Of course, all this really does is confirm Calvinball rules: the definitions change exactly as often as it takes to get the outcome someone in power wants. When “ownership” needed to mean any ByteDance stake to justify a ban, it meant that. Now that the administration wants TikTok back on its phones, “ownership” apparently means “controlling stake,” and 19.9% doesn’t count.

The real lesson here appears to be that the earlier concerns were exaggerated. Josh Hawley and Congress wanted to get headlines about how they were “taking on China” and “big tech” more than they wanted to write a clear law. They had a moral panic about one specific app, dressed up in national security language, and now that the political winds have shifted, the Office of Legal Counsel is left doing contortions to make the text say what the moment requires.

Advertisement

Filed Under: donald trump, josh hawley, olc, ownership, tiktok ban

Companies: bytedance, tiktok

Source link

Advertisement
Continue Reading

Tech

What matters more to an employer when hiring, AI or people skills?

Published

on

Murugan Anandarajan and Cuneyt Gozu of Drexel University explore the characteristics that make a job applicant more attractive in 2026.

If you spend a few minutes talking with college students about their career prospects, one topic is at the top of their minds: Will their jobs be displaced by a bot?

Their concern makes sense. Universities are integrating artificial intelligence into their curricula, and employers are adopting AI at a rapid pace. So most students assume that AI skills are now key to getting an entry level job.

But what if employers are changing their candidate criteria just as quickly?

Advertisement

As scholars who research how technology is changing the workplace, we have long asked what employers value when hiring new talent. The question grew out of our conversations with both employers and students about internship experiences.

Year after year, employers emphasise communication, professionalism and other workplace skills as the areas where interns most need to improve. Students tend to focus on building their technical skills instead.

To determine whether this disconnect extended beyond internships, we surveyed more than 600 employers across a range of industries and organisation sizes. We asked them which skills and qualities matter most when hiring recent graduates and early-career professionals.

Their answer might surprise you – they consistently ranked ‘soft skills’ over technical know-how when evaluating applications.

Advertisement

Learning to grow with the job

We had expected employers to place AI and technical skills at the top of their list. Instead, our survey told a more nuanced story. Employers clearly valued those skills, but they consistently ranked even higher qualities such as interpersonal skills and willingness to learn.

No matter the size of the organisation, the results were similar. The rankings showed what employers valued, and their comments explained their reasoning. One theme surfaced repeatedly – employers want to hire people who they believe will grow with their organisations.

This doesn’t mean technical abilities are unimportant. Rather, those were seen as job-specific skills that could be taught on the job – which is much more easily done when new hires already have soft skills like dependability, professionalism and a willingness to learn.

A human resources manager at a mid-size healthcare employer captured this idea well: “Computer skills, excellent customer service and dependability are of utmost importance. The rest, we can train.”

Advertisement

A small medical device company made a similar point: While technical positions require a base level of competence, attitude and adaptability also matter because employees often need to take on varied responsibilities rather than staying in ‘one lane’.

Another respondent, from a small nonprofit, emphasised curiosity, adaptability and the ability to learn new skills quickly. In a small organisation, the employer explained, employees need to be willing to grow along with the organisation.

Together, these comments suggest that employers aren’t choosing between technical abilities and workplace qualities. Candidates may still need a technical foundation, but employers also want evidence that they can learn, adjust and apply what they know in changing circumstances.

Can we talk?

Communication was another recurring concern.

Advertisement

A hiring manager at a large manufacturer described the challenge of finding candidates who were comfortable with face-to-face and telephone conversations. Similarly, an executive from a small professional services firm said poor writing was a recurring reason for rejecting applications, and he cited communication with clients and colleagues as a growing challenge.

While our survey doesn’t show whether communication skills have declined over time, the consistency of these comments suggests that employers see this as a persistent challenge when hiring recent graduates. One respondent shared that many young graduates were “not equipped with great communication skills” and had trouble following through on commitments.

These comments suggest that employers see communication as more than just speaking and writing. They also associate it with professionalism and reliability.

The comments also helped explain why internships and training programmes remain valuable to employers. A respondent from a small professional services firm said they primarily hired former trainees, adding: “We know their skills, motivation and how they fit within our firm.”

Advertisement

These experiences allow employers to observe more than a candidate’s academic record or list of skills. As part of our ongoing conversations with employers, one supervisor described an intern who completed several complex projects ahead of schedule while adapting quickly to a new corporate environment. The supervisor was especially impressed by the student’s communication skills and eagerness to learn.

As for AI, the message from employers was clear: Knowing how to use a technology matters, but so does knowing when and how to use it responsibly.

As one respondent wrote: “If AI is used well in application materials, it’s fine. Most of the time, we only notice because it’s being used so poorly, which is unacceptable.”

We encountered a similar issue while serving on a recent search committee. Several applicants appeared to have used generative AI to prepare their cover letters, but they left out references to other institutions or positions. Instead of strengthening their applications, these errors called into question the applicants’ judgement, attention to detail and the accuracy of their materials.

Advertisement

The hiring process through an employer’s eyes

Reading these comments changed the way we thought about hiring. We saw how students and employers often approach the hiring process from different perspectives.

Students, understandably, focus on getting the job. They build strong résumés, prepare for interviews and learn how to use AI. Those efforts matter because employers expect graduates to arrive ready to contribute.

Employers, however, are asking a different question: Is this someone who will succeed here?

Every interaction helps them answer that question. An email can show how clearly and professionally a candidate communicates. An interview may reveal how someone listens and responds to an unfamiliar question. An internship or training programme demonstrates whether a student follows through, works well with others and responds to feedback.

Advertisement

Viewed this way, the survey results become easier to understand. Employers aren’t simply evaluating what candidates already know. They’re looking for signs of how candidates will perform, learn and develop after joining the organisation.

Lessons for students and universities

For students, the message isn’t that they must choose between technical and workplace capabilities. They need both.

Learning to use AI and other workplace technologies remains important, but students also need to prove they can communicate clearly, work effectively with others, exercise judgement and follow through on their commitments.

This means universities need to do more than add AI-themed courses and majors to their curricula. They should give students repeated opportunities to demonstrate how they apply their knowledge, respond to feedback and contribute to a team. Internships, traineeships and client projects allow students to refine these qualities before entering the workforce.

Advertisement

Career advising can reinforce this strategy by helping students explain not only which skills they possess, but where and how they have used them.

We began with a simple question: Has AI changed what employers value when hiring recent graduates?

The answer was more nuanced than we expected. AI is changing how work gets done, but it hasn’t fundamentally changed what employers look for in the people they hire. They still want graduates who can use technology effectively while demonstrating the judgement, professionalism and adaptability needed to grow with the organisation.

After all, AI may be changing the workplace, but employers are still hiring people, not algorithms.

Advertisement

The Conversation

By Murugan Anandarajan and Cuneyt Gozu

Dr Murugan Anandarajan is a professor of decision sciences and management information systems at Drexel University. His research focuses on cybercrime, unstructured data analysis, business analytics and the strategic management of information systems. He teaches courses in text mining, qualitative research methods and disruptive technologies. 

Cuneyt Gozu is an associate clinical professor of organisational behaviour at Drexel University’s LeBow College of Business and serves as academic director of the LeBow Career Readiness Center. He teaches undergraduate and graduate courses in leadership, organisational behaviour, change management and career development. His research and applied work focus on college hiring trends, career readiness, leadership development, the future of work, and issues of power, influence, and emotional intelligence in organisational settings. 

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Advertisement

Source link

Continue Reading

Tech

Agentic coding goes hands-free as OpenAI brings GPT-Live’s full duplex voice control to Codex and ChatGPT on the desktop

Published

on

Two weeks after debuting its more naturalistic GPT-Live audio AI model with full-duplex capabilities (listening and speaking at the same time), OpenAI is bringing it directly into developer workflows.

The company announced that GPT-Live now powers the ChatGPT desktop application on macOS and Windows, integrating directly with agentic systems like Codex and ChatGPT Work (which are separate experiences available in the ChatGPT desktop app).

When OpenAI initially launched GPT-Live on July 8, 2026, it introduced a continuous audio model capable of listening and speaking simultaneously—eliminating rigid turn-taking while delegating complex reasoning to background models like GPT-5.5.

Today’s release expands that conversational layer to technical tasks, enabling software engineers to orchestrate multi-threaded coding jobs, review pull requests, and debug applications using natural voice commands.

Advertisement

As such, it could usher in a new era of “hands free” software development and even live, in-person group coding parties for the more than 10 million weekly active users across Codex and ChatGPT Work. Codex, of course, is the name given to OpenAI’s models and harness focused on coding, but which the company has this year expanded into a more general productivity platform. An OpenAI spokesperson told VentureBeat this is the first time voice activation

OpenAI posted a promotional video showing some of its employees, Codex developer experience engineer Jason Liu and Codex technical staffer Guinness Chen, speaking to the same ChatGPT desktop app session in the same room, each issuing different instructions and conversing with the same model.

New capabilities unlocked

At its core, this integration relies on decoupling the real-time voice layer from the underlying execution engines.

Advertisement

While GPT-Live maintains fluid conversation—inserting natural verbal acknowledgments like “got it” without interrupting the user—it passes heavy computational workloads to background reasoning models.

On macOS, the desktop application incorporates “Appshots” and screen context features, allowing ChatGPT Voice to analyze the frontmost window alongside local files, codebase structures, and active plugins.

This architecture creates a pair-programming dynamic where developers talk through problems conversationally while agents execute tasks asynchronously.

Rather than manually stopping coding sessions to type detailed instructions or switch windows, developers direct the system hands-free.

Advertisement

The full-duplex engine dynamically decides when to speak, pause, or invoke tools, maintaining conversational state even as background agents process complex code modifications.

Directing coding and complex builds with your voice alone

The central operational capability in this update centers on multi-task execution across Codex and ChatGPT Work environments.

Software engineers can initiate multiple concurrent task threads from a single spoken prompt. For instance, a developer preparing to ship a feature can instruct the system to investigate an open authentication bug, review a pending API migration pull request, and generate missing unit tests simultaneously.

The desktop application coordinates these actions across disparate contexts, tracing issues through Slack conversations, GitHub repositories, and local codebases.

Advertisement

Developers can also verbally convert design mockups into working code, splitting tasks across frontend, backend, and testing layers.

With support for multi-folder projects (build 26.715) and remote execution via iOS, engineers can check task progress, answer agent prompts, and redirect active jobs without switching applications or managing individual processes line by line.

Proprietary license

OpenAI’s voice-enabled desktop release operates under a proprietary, commercial enterprise model. Access is restricted to paid subscribers across Plus, Pro, Business, Enterprise, and Education plans.

For individual developers and corporate engineering departments, this commercial structure means the model weights, voice processing pipelines, and agent state architectures remain fully closed.

Advertisement

Organizations cannot modify or self-host the underlying systems. Furthermore, tasks initiated via ChatGPT Voice consume standard usage allocations directly from existing Codex and ChatGPT Work plan quotas, treating voice-triggered actions identically to standard agentic workloads.

Community reactions

Developer communities immediately noted the implications of bringing continuous full-duplex voice to autonomous coding workflows.

Reacting to the build 26.715 release announcement—which details voice integration and multi-folder project support—AI Insider journalist @ChrisGPT noted on X: “Today OpenAI will release voice and remote guidance for codex ! One step closer to personal AGI”.

Early technical feedback highlights widespread enthusiasm for orchestrating complex agentic tasks hands-free, particularly when stepping away from the workstation or managing build pipelines remotely.

Advertisement

Source link

Continue Reading

Trending

Copyright © 2025