Frankramspott/Getty Images

We may receive a commission on purchases made from links.

You wake up in the morning and check with your weather app, which tells you that there is no snow today. You expect an easy commute. Ten minutes on the road, however, the flakes start to fall. The problem with modern weather apps is the reliance on algorithms with no real-time human oversight, which sometimes leads to inaccurate weather predictions. All apps can be susceptible to this, even if they have some of the best weather service features.

While you can tune into your local station to get a meteorologist’s opinion, there is one piece of tech that doesn’t require turning on the TV or going to the weather app on your smartphone. That is a Galileo thermometer. This retro thermometer, invented by astronomer Galileo Galilei, is great for showing local temperature in your area as well as predicting changes. These are available to buy at online retailers or in some physical stores.

Aside from offering an alternative to local weather prediction, Galileo thermometers are quite pretty to have on display and offer a nice, screen-free solution. Since staring at our phones too much can become addicting and lead to high levels of anxiety and depression, this is a way to remove yourself from constantly looking at your phone.

How a Galileo thermometer works

Hmvart/Getty Images

A Galileo thermometer works by taking advantage of how water’s density changes with temperature. Inside the sealed glass tube is water and several colorful glass bubbles, each tagged with a different temperature. The tags change each bubble’s weight, making every bubble a tiny bit denser or lighter than the others. As the surrounding air warms or cools, the water inside the thermometer expands or contracts, which changes its density. When that happens, some bubbles sink while others float. The lowest bubble that’s still floating matches the water’s density and shows the current temperature. Because rising or falling air pressure often brings temperature changes with it, watching these subtle shifts can give a rough clue about where your local weather might be headed.

Laurahilchey/Getty

Amazon offers such a thermometer also equipped with storm glass and an analog clock for $44 at the time of writing. It’s a neat full-service weather station. The Galileo thermometer itself shows the local temperature. The storm glass enhances the display of potential weather forecasting based on the changing temperature. Finally, it has a hygrometer for local humidity readings. The Galileo thermometer is a pretty piece to set out either at home or in your office. If you don’t want to spend $44, you can wait around for some of the best days to buy tech on Amazon to try to snag a better deal.

Why weather apps can be unreliable

Alex Photo Stock/Shutterstock

While you might consider your weather app essential for daily activities, and perhaps even among the great essential apps if you are an outdoor enthusiast, the problem is that they are not always reliable. This is especially true in areas that can have unpredictable weather patterns, like lake effect snow around the Great Lakes or the way the Rocky Mountains affect weather patterns ithose states.

Weather apps don’t have meteorological oversight to constantly check and update the data. Instead, they combine data from satellites, radar systems, and ground weather stations. The data gathered from these sources is then put together into forecasting models for the app. These apps may pull from various different computer weather models, which explains why two different weather apps can show two different forecasts for the day.

While these weather models and data processing systems are certainly advanced, they still rely on typical patterns. Subtle changes may not be accounted for and the way local geography impacts local areas is not always reflected. That is why your weather app may be wrong, and one of the benefits the Galileo thermometer has over it. The Galileo thermometer shows the weather as it happens, with the bubbles inside moving to forecast what you can expect in the near future. Though the Galileo thermometer has its limits in terms of wide-range, long-term forecasting, it is still a handy tool to have at your disposal.

Source

Elvard project/Shutterstock

Well, it seems Google is up to some more shenanigans with the upcoming arrival of the Pixel 10a, its next budget-friendly Android phone. While our review of the Pixel 10 praised what Google is doing with its smartphones, the company still seems intent on trying to win over as many Apple iPhone users as possible. That’s because on the latest store page for the Pixel 10a, Google actually advertises being able to use FaceTime — Apple’s exclusive video chatting functionality — with its latest phone.

Now, it’s important to note that this isn’t the first time that we have seen Google doing something like this. In the past, Reddit users spotted similar claims on the Pixel 10’s store page, and the same advertisement bit actually shows up on the main store page for the Pixel lineup. So, what’s the big deal? Does the Pixel lineup offer some kind of special deal with Apple that gives users access to FaceTime on Android?

No. It does not. While it might sound like Google is saying that FaceTime is more readily available on Pixel phones, you have to look at the fine print to truly understand what that statement means. Because, instead of the Pixel actually offering any kind of extended support for FaceTime, Google is simply riding the addition of FaceTime links in iOS 15, which allows Apple users to invite both Android and Windows devices to FaceTime calls by creating a call group and then sending the link to the other users.

What Google’s FaceTime advertisement really means

Being able to join FaceTime calls on Android is absolutely a handy functionality to have, and just one way that Apple has been working to make iPhones and its other products work more seamlessly with Android and Windows devices. However, there’s nothing different about what the Pixel is offering here. Instead, it seems Google might just be trying to add in some extra buzzwords to help entice new users to pick up its phones. There’s nothing inherently wrong with Google outlining that you’ll still be able to chat with FaceTime users on Pixel. However, the exact wording has certainly led to some confusion with consumers.

Of course, Google’s own response to FaceTime, Duo, arrived on iPhone and Android several years ago, but slowly died off as the service was pushed into Google Meet. That’s likely because FaceTime works so seamlessly with devices, with some on Reddit even calling it “phone calling 2.0.” As such, this ongoing popularity and ease-of-use is probably why Google is using it as a selling point for its devices. Another confusing part of the equation, too, is the fact that Google actually made AirDrop work with its Android phones, so it makes sense that some might believe they’d somehow cracked the code or made a deal with Apple for FaceTime support. But that remains unlikely.

Source

rukawajung/Shutterstock

We may receive a commission on purchases made from links.

Driving a car is a lot of responsibility. On the road, there is no shortage of laws to adhere to and plenty of other drivers to contend with. While it’s up to everyone to keep the avenues and highways safe, it’s not a job all drivers take seriously. There are plenty of dangerous people behind the wheel, and the only individual who can defend you, yours, and your vehicle from the brake-checkers and speed demons of the world is you. 

Fortunately, there are also plenty of car-friendly USB port gadgets on the market, and many are designed with safety in mind. Whether you’re looking for front and rear dash cams that never stop recording, or a plug-and-play LED light in case of emergencies, you can make all your four-wheel excursions safer and more efficient with any of the five devices we selected.  

When choosing items, we did our best to stick with top-rated devices with bountiful user feedback and hands-on reviews from trusted publications. We also made sure to include a couple of budget-friendly devices that folks with nothing more than $15 can reap the benefits of. 

70mai A810 Dash Cam

Everyone should own a multi-view dash cam system, and frankly, they should be built into all new road-bound vehicles. But until that day comes (if it ever does), we’ll be forced to invest in our camera tech. Fortunately, cam combos like the 70mai A810 are readily available, and at a price most folks will be able to wrap their heads around.

The entire A810 rig runs off USB-A and DC power, and comes with a front-facing cam that delivers 4K recordings and a 146-degree field of view (FOV). The back cam records in 1080p HD, has a 130-degree FOV, and plugs into the side of the front cam for power. The system stores all footage to a microSD and accepts cards up to 256GB.

On Amazon, the A810 combo scored a 4.4 out of 5 stars, based on nearly 1,600 reviews. Most users were pleased with the overall video quality and easy setup. While the rear cam could do with some processing improvements — and perhaps a more conventional USB connection down the line — the 70mai A810 is well worth the investment. Plus, it’s often on sale for less than its $180 list price.

Ecoangel LED Road Flares

No one ever wants to be in a position where firing up a road flare is necessary, but it’s better to be safe than sorry. To that end, we recommend the Ecoangel LED Road Flares 3-Pack, a versatile set of disc-shaped emergency lights with nine different flash modes and flashlight capabilities. Each disc contains a 650-milliamp lithium-ion battery that recharges via USB, and all three units are waterproof.

Perfect for flagging down the AAA tow that’s coming to rescue you, the road flares can be programmed to flash intermittently, strobe repeatedly, or stay solid. When it’s time to recharge, simply use the provided USB cable to connect to your vehicle, and wait until the flare indicator switches from red to green (indicative of a full charge).

Amazon shoppers gave this bundle 4.8 out of 5, based on over 4,300 reviews. Most folks love how well the lights work and how simple they are to use. Others praised the durability of the product, as well as the magnetization that lets you stick them to certain parts of your vehicle. Ecoangel was even kind enough to provide a carrying case for all three flares.

AstroAI L7 Tire Inflator

A tire repair tool is another gadget every driver should have, and we think the AstroAI L7 Tire Inflator is worth every penny. It scored a 4.3 out of 5 stars on Amazon, and that’s based on over 10,300 user reviews. The AstroAI L7 has a 4,000-milliamp battery, which the manufacturer claims is enough power to inflate up to eight tires (195/65R15 size) from 30 to 36 PSI.

Conveniently, it even pulls double duty as a backup charger for phones and other tech products. The USB ports are located on the bottom of the gauge, and AstroAI recommends recharging every three months for optimal performance. With four vehicle modes to choose from and a custom mode that lets you set your own PSI value (3-150 PSI), the L7 probably won’t be a tool you use often — but it’s one you’ll be glad to have should a tire decide to deflate.

The tire inflator comes with everything you need in the box, including a USB charging cord, air nozzles, a needle valve adapter, and a Presta to Schrader adapter. We read a few reports of folks who ran out of battery faster than anticipated, so definitely make sure you stick to that three-month recharge schedule.

RunningSnail Emergency Radio

A portable radio may not be at the top of your list of cheap car gadgets to invest in, but this RunningSnail Emergency Radio shouldn’t be ignored. Priced at $27, it’s inexpensive even when it’s not on sale. What you’re getting is a 2,000-milliamp hub with AM/FM/NOAA tuning, an integrated LED flashlight, and three ways to charge the battery: USB, solar, and a manual hand crank.

Measuring 5 inches tall and 2.4 inches wide, the RunningSnail is small enough to fit in a pocket, making it easy to switch the device between vehicles. If push comes to shove, and your phone is minutes from dying, you’ll even be able to use the radio’s USB port to top off mobile devices. That said, 2,000 milliamps isn’t a ton of power, so it’s still a good idea to carry around one of the best portable chargers.

Amazon shoppers gave this RunningSnail radio 4.5 out of 5 stars, and that was based on over 22,750 reviews. People love the reliability of the device when the power’s out, as well as the audio quality of the weather stations it’s able to pick up. Some users complained that the battery doesn’t hold a charge, while others had no issues. To play it safe, we recommend unboxing and giving it a test run.

Ztylus Stinger Escape Tool

Car escape tools aren’t the friendliest-looking gadgets, but a tool like the Ztylus Stinger Escape Tool should be an essential part of your vehicle’s emergency kit. The window-breaking mechanism is spring-loaded; all you have to do is press the Stinger up against one of your tempered glass side windows and press the trigger. We also appreciate the inclusion of a seatbelt cutter.

One of our favorite features is the DC connector the device uses. When plugged into your car’s 12-volt DC port, the Stinger doubles as a dual USB charger for phones and other portable tech. Amazon shoppers gave it a 4.6 out of 5 stars, based on over 1,850 reviews, and most folks were pleased with how well the Stinger performs. Others weren’t so keen on how long it takes the USB-A ports to charge a phone.

Even if you don’t want to keep the Stinger plugged into your vehicle’s DC port, it’s small enough to be tucked away in a glovebox or center console. Just make sure you know exactly where it is at all times.

Methodology

Anyaberkut/Getty Images

Vehicle-friendly emergency gadgets may not be as fun or feature-packed as the consumer tech products we usually cover, but these types of tools exist for several reasons. Saving lives and keeping people safe is at the top of the priority list, and safety and convenience are exactly what we had in mind when choosing these five products. We also made sure to stick with reliable brands and parsed both user feedback and professional reviews for pros, cons, and other critical feedback. Lastly, we did our best to include a couple of products that even the strictest of safety budgets can afford.

Source

The Metropolitan Police’s rapid “unchecked” expansion of live facial-recognition (LFR) technology is taking place without clear legal authority and minimal public accountability, says Green London Assembly member Zoë Garbett in a call for the force to halt its deployments of the controversial technology.

Made during an ongoing government consultation on a legal framework for the technology, Garbett’s call for the force to immediately halt its deployments of LFR is informed by concerns around its disproportionate effects on Black and brown communities, a lack of specific legal powers dictating how police can use the tech, and the Met’s opacity around the true costs of deploying.

Garbett’s intervention also comes as the High Court is considering the lawfulness of the Met’s approach to LFR, and whether it has effective safeguards or constraints in place to protect people’s human rights from the biometric surveillance being conducted.

“Live facial-recognition technology subjects everyone to constant surveillance, which goes against the democratic principle that people should not be monitored unless there is suspicion of wrongdoing,” said Garbett, adding that there have already been instances of “real harm” in children being wrongly placed on watchlists, and the disproportionate targeting and misidentification of Black Londoners.

“These invasive tools allow the police to monitor the daily lives of Londoners, entirely unregulated and without any safeguards. The Met repeatedly claim that live facial recognition is a success, yet they continue to withhold the data required to scrutinise those claims.

“It makes no sense for the home secretary to announce the expansion of live facial recognition at the same time as running a government consultation on the use of this technology. This expansion is especially concerning given that there is still no specific law authorising the use of this technology.”

Highlighting in a corresponding report how facial-recognition technology “flips the presumption of innocence” by turning public spaces into an “identification parade”, Garbett also outlined ways in which both the Met and the Home Office can make its use safer in lieu of a full-blown ban.

This includes creating primary legislation with “strict controls” that limits LFR to the most serious crimes and bans its use by non-law enforcement public authorities or the private sector; and openly publishing deployment assessments so that watchlist creation, location choice and tactical decisions are publicly available for Londoners to review.

On watchlist creation specifically, Garbett dismissed the police claim that LFR is a “precise” tool, highlighting how nearly every watchlist used is larger than the one preceding it.

Highlighting how the number of faces being scanned by the Met is “increasing at a near exponential rate”, Garbett likened the forces watchlist tactics to a “fishing trawler” that it keeps adding to so it can find people.

“Data suggests that rather than making a new unique watchlist for each deployment based on the likelihood of people being in the area of the deployment, it seems from the outside that the MPS is just adding additional people on to a base watchlist [it has],” she said.

Garbett also called on the Met to publish the true financial and operational costs of all LFR deployments, arguing that the force has not only failed to provide a compelling business case for the technology, but is actively obfuscating this information.

“The MPS has a history of a lack of transparency. This is perhaps best summarised by Baroness Casey in her review of the MPS where she said, ‘The Met itself sees scrutiny as an intrusion. This is both short-sighted and unethical. As a public body with powers over the public it needs to be transparent to Londoners for its actions to earn their trust, confidence and respect’,” said Garbett.

She added that while freedom of information requests returned in mid-2023 revealed that, up until that point, the force had spent £500,000 on the tech, without up-to-date reliable figures, it is impossible to verify the Met’s claims that it is delivering a greater impact on public safety through LFR.

“The NHS wouldn’t be able to roll out a new treatment without being able to prove it was worthwhile and effective, but it seems that the police operate under their own rules and seemingly answer to no one,” said Garbett.

Computer Weekly contacted the Met about Garbett’s report. A spokesperson said that LFR “has taken more than 1,700 dangerous offenders off the streets since the start of 2024, including those wanted for serious offences, such as violence against women and girls. This success has meant 85% Londoners support our use of the technology to keep them safe.

“It has been deployed across all 32 boroughs in London, with each use carefully planned to ensure we are deploying to areas where there is the greatest threat to public safety. A hearing into our use of live facial recognition has taken place and we look forward to receiving the High Court’s decision in due course. We remain confident our use of LFR is lawful and follows the policy which is published online.”

A lack of meaningful consultation so far

While the use of LFR by police – beginning with the Met’s deployment at Notting Hill Carnival in August 2016 – has already ramped up massively in recent years, there has so far been minimal public debate or consultation, with the Home Office claiming for years that there is already “comprehensive” legal framework in place.

The lack of meaningful engagement with the public by police and government over facial recognition is reflected in Garbett’s report. She highlights, for examples, that Newham Council unanimously passed a motion in January 2023 to suspend the use of LFR throughout the borough until biometric and anti-discrimination safeguards are in place.

While the motion highlighted the potential of LFR to “exacerbate racist outcomes in policing” – particularly in Newham, the most ethnically diverse of all local authorities in England and Wales – both the Met and the Home Office said that they would press forward with the deployments anyway.

“Since that motion was passed, LFR has been used 31 times in Newham by the MPS,” said Garbett.

On the deployment of permanent LFR cameras mounted to street furniture in Croydon, Garbett added while the Met promised it would consult with the local community, councillors from there are have told her the force did not follow through with this consultation.

The technology was similarly rolled out in Lewisham without meaningful consultation, despite the Met’s claims to the contrary.

However, in December 2025, the Home Office launched a 10-week consultation on the use of LFR by UK police, allowing interested parties and members of the public to share their views on how the controversial technology should be regulated.

The department has said that although a “patchwork” legal framework for police facial recognition exists (including for the increasing use of the retrospective and “operator-initiated” versions of the technology), it does not give police themselves the confidence to “use it at significantly greater scale…nor does it consistently give the public the confidence that it will be used responsibly”.

It added that the current rules governing police LFR use are “complicated and difficult to understand”, and that an ordinary member of the public would be required to read four pieces of legislation, police national guidance documents and a range of detailed legal or data protection documents from individual forces to fully understand the basis for LFR use on their high streets.

Consultation responses

In a section on how people can respond to the Home Office’s LFR consultation, Garbett urged people to call for its ban, adding that further protections in lieu one could include requiring a warrant to be placed on a watchlist, and limiting it to “the most serious and urgent crime purposes”.

She noted that, as it stands, the Met has not used LFR to make any terror-related arrests, with the most common offence being variations on theft or court order breaches

“In a recent press release, the lead example the MPS give for how they have used LFR is using it to arrest a 36-year-old woman who was wanted for failing to appear at court for an assault in 2004 when they were probably 15 years old,” she said. “The public might feel differently about LFR if they knew it was being used on cases such as these.”

On the permanent installation of LFR cameras in Croydon, Garbett added that while the police have said they are only switched on when an operation is taking place, “there is still the potential for 24/7 monitoring, with Londoners unable to tell if the cameras are operational or not. This makes the feeling of being under surveillance in London feel routine and begins to be a slippery slope to preventative policing and a blurry line between safety and social control.”  

Garbett concluded that the rapid deployment of LFR must stop while safeguards are in place to protect people’s rights: “I urge everyone to respond to the government consultation and use the guide I’ve prepared to make sure we have a say in how this technology is used going forward.”

Computer Weekly contacted the Home Office about the contents of Garbett’s report and its decision to massively expand facial-recognition deployments before concluding its consultation.

“Facial recognition is a crucial tool that helps the police locate suspects and those wanted by the courts. In the past two years alone, it has helped the Metropolitan Police locate and arrest more than 1,700 offenders, including rapists, domestic abusers and sex offenders,” said a spokesperson.

“The Home Secretary has announced plans to roll out facial recognition across the country. The number of live facial recognition vans will triple, with 50 vans available to every police force in England and Wales.”

Source

The total number of common vulnerabilities and exposures (CVEs) disclosed in 2026 is set to romp past the 50,000 mark in 2026 and may plausibly run as high as six figures for the first time ever, according to the Forum of Incident Response and Security Teams’ (First’s) annual Vulnerability report.

In its latest set of predictions, First said that this year, the upper bounds of its 90% confidence interval in fact approaches 118,000 CVEs, and according to the data, realistic scenarios suggest 70,000 to 100,000 disclosed vulnerabilities are “entirely possible”. The median figure for 2026, it said, would most likely be around 59,000.

First said that whatever the figure turns out to be, it underscored an “urgent need” for organisations to both scale their security ops and strategically prioritise their vulnerability response and patching practices.

“The question organisations need to ask right now is: are my people and processes ready to handle this volume, and am I prioritising the vulnerabilities that actually put my data at risk?” said Éireann Leverett, first liaison and lead member of First’s Vulnerability Forecasting Team.

“Our forecast allows defenders to stop reacting to every new CVE and start making strategic decisions about where to focus limited resources before attackers exploit the gaps.”

The 50,000 vulnerability question

In its 2025 report, First said that the higher end of its predicted range topped out at 50,000 CVEs – the number its analysts expect to comfortably exceed this year. This was partly due to the rapid adoption of open source software (OSS) and the use of AI tools both in vulnerability discovery. During the course of the year, the emergence of the vibecoding phenomenon likely also had an impact.

In the event, First’s prediction was bang on, Leverett revealed, tipping over the upper confidence mark on 31 December 2025 for a final total of 49,972 observed CVEs, just 28 short of the magic number.

However, ideally, the upper confidence point would fall somewhere in 2026, with the median confidence point falling on New Year’s Eve, and as a result, First has reviewed its approaches and methodology going forward. Whether or not this means its 2026 forecast will be even more accurate remains to be seen.

“[Our] new method of forecasting…allows for asymmetric confidence intervals. This means we are taking into account that the publication number is more likely to exceed last year than be less than last year,” Leverett told Computer Weekly.

“So while we expect the number to be closer to 60,000, there is a 10% chance it exceed 118,000. Most of this is just statistics, but there is also discussion about emerging technologies and how they might stretch the range of possible numbers, which meant we were more comfortable publishing the results of this modelled outcome than some others.”

Next steps

While at first glance First’s annual CVE report might seem just an interesting statistical marker, the forecast serves as a potentially critical planning tool for the security sector when it comes to planning patching capacity, writing coordinated disclosures, or developing new detection signatures for SIEM, EDR or IDS platforms.

“Much like a city planner considering population growth before commissioning new infrastructure, security teams benefit from understanding the likely volume and shape of vulnerabilities they will need to process,” said Leverett. “The difference between preparing for 30,000 vulnerabilities and 100,000 is not merely operational, it’s strategic.”

Whether they end up facing 50,000 or 100,000 CVEs, and keeping in mind that not every flaw will affect every business, security leaders at end-user organisations can start the work to get out in front of the problem right now.

A strong jumping off point is to assess whether the organisation has the people, processes and capacity to handle so many issues. A well-prepared CISO will have prepared for the median forecast but will also have built contingency plans for the higher-volume scenarios.

Security pros also need to master the art of ruthless prioritisation, focusing on the flaws that pose the greatest risk to their specific IT estates, and not just those with the most critical CVSS numbers.

Finally, leaders should leverage external vulnerability forecasts alongside their own asset inventories to make vendor- and product-specific preparations.

“No company can solve vulnerabilities and cyber security in isolation. The organisations that recover fastest are the ones with trusted networks already in place, sharing threat intelligence and coordinating response before a crisis hits,” said First CEO Chris Gibson.

Source

The National Audit Office (NAO) has refused to endorse the audited accounts of the College of Policing for a second year running, as the policing organisation continues to recover from serious failures in an IT project that left it unable to properly manage its finances.

The accounting watchdog said that, although there had been no new financial issues, it “disclaimed” the college’s latest 2024–25 accounts because of the continuing “fallout” from financial reporting problems that resulted from a problematic IT project.

“We were not able to provide a complete opinion on the opening position or in-year transactions for 2024–25, both of which are heavily derived from the closing position of the prior year,” it added.

The College of Policing, the professional body for policing in England and Wales, ran into difficulties when it replaced its SAP-based accounting systems with the Home Office’s Oracle-based Metis accounting system in October 2023.

It transferred its payroll systems from CGI UK IT limited to a new supplier, Shared Services Connected Limited (SCCL), on the same day.

‘Significant concerns’

The move left the college unable to produce accurate figures for financial transactions, leading to “significant concerns” about the integrity of the college’s financial records, which contributed to an overspend of £1.3m.

College CEO Andy Marsh said in a statement that it had now addressed its previous financial problems.

“We have introduced stringent new procedures to stabilise our financial processes and made further progress by enhancing the college’s financial controls, including building greater resilience and expertise at board, executive and operational levels,” he added.

Computer Weekly reported in July 2025 that the NAO found the college failed to manage the risks of the project, and did not address known defects before going live with its new accounting systems.

There was a failure to segregate financial report data held by the Home Office and the college, and the problem remained unresolved during testing and “go-live”, leading to “potential inaccuracies” in financial reporting, the NAO found. Issues with data conversion and migration were also not resolved, creating further risks to the integrity of financial data.

The college had failed to check on a “line by line” basis that the transactions on the SAP systems had been accurately and completely transferred to Metis, an outsourced service shared with other government departments, and was unable to obtain “a significant amount of information” required from its 2023–24 financial audit.

Contract issues

The Home Office’s contact with SCCL did not require the service provider to hand over the payroll information the college needed for its 2023–24 audit, delaying the information required by auditors to complete their work by four months.

The problems were exacerbated because the college lacked people with the right technical and financial skills. It had only one member of staff with knowledge of the SAP accounting system, who went on an extended leave of absence, and the board member overseeing the accounting team was not a qualified accountant.

Marsh added that the National Audit Office acknowledged the improvements the college has made this year.

“The auditor stated that: ‘For 2024 to 2025, the college has successfully produced a set of auditable financial statements, which is a significant achievement from a difficult starting position.’ This progress represents a crucial step in the college’s financial recovery and is a notable achievement given our challenging starting point,” he said.

“In 2023, we encountered major challenges with our accounts, caused by the introduction of a new finance and HR system. The auditor’s disclaimer on the 2024 to 2025 accounts relates solely to these previous financial problems, which have now been addressed.”

The college described this year’s audit as the first step in a three-year audit recovery plan. It remained “on trajectory” to restore a fully unqualified audit opinion in 2026–27, according to its published accounts.

The college said it had conducted a lessons learned exercise and undertaken best practices training with the NAO, in addition to appointing a chief financial officer and director of delivery.

It addressed 40 technical and systems issues identified by the NAO, and is continuing to work through the list with the support of the Home Office.

The college also worked with SSCL to ensure that previous problems with not having “timely access” to audit information – particularly payroll – were not repeated.

Source

The European Commission (EC) has preliminarily found that TikTok’s addictive design features violate the bloc’s Digital Services Act (DSA).

The preliminary decision outlines how addictive design features on the platform, such as infinite scroll and autoplay, are resulting in users going into “autopilot mode”, with the EC stating this may lead to “compulsive use”.

The DSA, which sets out rules for online services used by European citizens, is designed to strengthen consumer rights and consumer choice, while also minimising the risk of harm. The act also requires platforms to carry out a risk assessment of negative effects on children’s mental health and present it to the EC.

TikTok is one of the 17 companies defined as Very Large Online Platforms under the act, which means it has to comply with the most stringent rules of the DSA because the size of its user base means there is greater potential for systemic harms to occur.

In its ruling, the EC stated that TikTok had failed to implement reasonable and effective measures to mitigate risks from its addictive design features, arguing that minors and vulnerable adults are at particular risk of harm.

The EC’s investigation also revealed that TikTok’s risk assessment had not adequately addressed how its design features and dark patterns could cause harm to the physical and mental health of its users.

On the protective measures that are in place, including screen time management and parental control tools, the EC noted they “do not seem to effectively reduce the risks stemming from TikTok’s addictive design” due to being easy to dismiss or overlook.

“At this stage, the Commission considers that TikTok needs to change the basic design of its service. For instance, by disabling key addictive features such as ‘infinite scroll’ over time, implementing effective ‘screen time breaks’, including during the night, and adapting its recommender system,” it said.

European Union (EU) tech chief Henna Vikkunen told reporters that minors are more at risk because “they don’t have the same tools” to avoid compulsive behaviour.

The decision could force the app, which has more than one billion users globally, to make design changes to avoid penalties, the European Commission said.

If it fails to make the necessary changes, the app could face fines of up to 6% of annual revenue, which it reportedly hoped could reach $186bn last year. TikTok was also accused of breaking digital advertising rules over transparency in May 2025.

The ruling marks the first time the EC has taken a legal stance on the design features of a social media company, tackling what many online safety advocates and campaigners recognise as addictive design.

Responding to the EC’s preliminary findings, a TikTok spokesperson said they “present a categorically false and entirely meritless depiction of our platform, and we will take whatever steps are necessary to challenge these findings through every means available to us”.

However, many hope this could set a precedent for future action against recommender systems that amplify allegedly harmful and illegal content. In December, the first-ever fine for breaching the DSA was given to X, with a total of £104m.

These findings confirm what people have been saying for years: TikTok’s addictive design is not an accident, it’s a business model Ava Lee, People vs Big Tech

“These findings confirm what people have been saying for years: TikTok’s addictive design is not an accident, it’s a business model. We need [EC president] Ursula von der Leyen to stand up for European citizens and show the political leadership this moment calls for,” said Ava Lee, executive director of People vs Big Tech.

US states have already pursued a case against TikTok for its addictive design features, with the lawsuit alleging the product is damaging children’s mental health. 

The move by the EC comes at a time when concerns around the safety of social media platforms are growing across the continent.

France, for example, has voted on a social media ban for children under the age of 15, while Spain has proposed to criminalise algorithms that amplify illegal content. In his speech to the World Governance Summit in Dubai, Spanish prime minister Pedro Sanchez vowed to protect children “from the Digital Wild West”.

In January 2026, the UK’s House of Lords voted to back a social media ban for under-16s by 261 votes to 150, with the government launching a national consultation to discuss next steps for online safety and digital well-being.

Von der Leyen has expressed support for an EU-wide age limit, following Australia being the first country in the world to ban under-16s from accessing social media in December last year.

“Amidst current discussions of restrictions on children’s access to social media platforms, governments must remember they also have a duty to protect children’s right to participate in the digital world,” said Lisa Dittmer, Amnesty International researcher on children and young people’s digital rights.

“To do so, their focus must be on tackling the toxic design of leading social media platforms, including through effectively enforcing laws like the Digital Services Act,” she added.

Research by Amnesty International has previously found that despite risk mitigation measures announced by TikTok since 2024, the platform continues to expose vulnerable users to content that normalises self-harm, despair and suicidal ideation.  

The European Commission’s announcement of its preliminary findings comes just days after a report was published by the US government’s House Judiciary Committee, titled The foreign censorship threat, Part II: Europe’s decade-long campaign to censor the global internet and how it harms American speech in the United States.

The publication states that the European Commission has “pressured major social media platforms to change their global content moderation rules”, revealing growing discontent from the Trump administration towards EU tech regulation.

“For the first time, the European Commission is critically examining the recommender mechanisms through which platform operators manipulate the free choice of users,” said German MEP Alexandra Geese. 

“Coupled with high personalisation, this system distorts the idea of freedom online. I expect these recommender mechanisms to be scrutinised on other platforms, too. There are better algorithms for ensuring choice online. It’s not the users who want disinformation, hate and violence, but the platforms.”

Source

Following a report that Apple pushed back its plans to introduce an AI health coach subscription as part of its new features expected for iOS 27, Google is doing the exact opposite, introducing an AI health coach to Fitbit Premium users on iOS. According to a Fitbit community blog post, the health tracker is offering a Public Preview of its upcoming AI personal health coach for iOS users.

For now, only paid or trial Fitbit Premium subscribers with an up-to-date app and operating system, in addition to a few other requirements, can join the Public Preview. If they meet the criteria, they can take advantage of Google’s AI health coach, which has been available to try on Android devices for a few months now. Google says this 24/7 digital advisor can offer tips and guidance across three main points: fitness, sleep, and holistic health, which are all features the activity tracker already offers.

Here’s everything Fitbit’s AI coach will be tackling

According to Fitbit, the experience of using its AI health coach will be like going to the gym for the first time. You’ll talk with the AI about your goals and lifestyle, so it can understand how to provide you with the best data (as if you live an active life, if you’re trying to lose weight, get back on track after an injury, and so on).

The difference of this AI coach is that it’s expected to be highly adaptable, so if you had a poor night’s sleep, there’s something on your schedule, or priorities changed, it will give you guidance based on everything happening in your life. Besides that, Google says this AI coach looks at the “bigger picture,” and not only at daily goals, as the ultimate goal is to bring consistency to users’ health and fitness initiatives.

In an enterprise page, Google also highlights how the personal health coach can be great for preventive care, as it can understand and interpret your wellness data, in addition to overall fitness information and sleep consistency. The company says the health data isn’t used for Google Ads, and that users have total control over what to share with the AI personal coach and also what not. Still, not everybody is sold on these privacy claims.

What happened to Apple AI coach efforts?

José Adorno/BGR

Apple has reportedly been planning to introduce its AI health coach for at least over a year, as this feature was expected for last year’s WWDC 2025. While the company postponed the announcement, recent rumors indicate that it could be unveiled as soon as iOS 26.4, or even at the WWDC 2026 with the iOS 27 introduction.

However, with Bloomberg saying Apple scrapped its plans due to the AI health coach not meeting the standards for a new subscription, it seems the company might slowly integrate some of its ongoing efforts on the Health app in the upcoming updates. For example, Apple has reportedly made a studio in California to record medical-based information about health data, offering users guidance over their data.

Besides that, the company just started testing AI features for Apple Watch, with functions like Workout Buddy, and Apple Fitness+ AI dubbed programs to offer its catalog to a broader audience. At the end of the day, it seems Apple might continue to add value to its current subscriptions and features, even though it isn’t offering anything on top of what competitors already are. That said, we’ll discover more about its health-related plans in a couple of months, once WWDC 2026 kicks in.

Source

Digital twins – virtual representations of environments and dynamics of interest – can address a wide range of decision-making needs and opportunities, and expectations for the technology and related applications are high.

A study from Fortune Business Insights projects the market to grow from $24bn in 2025 to more than $250bn in 2032. Digital twins can support research, planning and operations across a wide range of application areas, such as biological systems, machines and infrastructures, industrial operations, communities and cities, and even simulations of global and geopolitical dynamics. Recent discussions have centred on their use for robotics and robotics management.

The versatility of digital twins is substantial, but hurdles exist that prevent them from reaching their full potential. Some dynamics are not fully captured, while other dynamics are difficult to address comprehensively. In some cases, artificial intelligence (AI) can reduce existing limitations, but use of AI can create its own problematic issues.

Injecting human behaviour

Machines and equipment can be modelled according to physical formulas and accumulated sensor data that capture real-time and real-world behaviour. The same is true for electricity and water networks, for example. These systems are complicated but can be modelled in theory.

Complicated systems behave in predictable ways. Complex systems, in contrast, will behave differently each time, in part because of human behaviour that can change according to many influences. Most digital twins tend to omit human behaviour, while others treat human behaviour as predictable placeholders – in a way, they mechanise human behaviour. But human behaviour and interactions are of crucial importance in simulating dynamics for digital twins for cities and urban environments – after all, that’s what cities and communities are ultimately created for.

Farzin Lotfi-Jam, assistant professor at Cornell University’s College of Architecture, studies the use of technologies to govern cities. He is the director of Cornell University’s Realtime Urbanism Lab, which “investigates the impacts of new technologies that virtualise cities and populations”.

He says: “The global proliferation of urban digital twin models compels a research agenda that investigates the intertwined social, political and technical dimensions of their development, from design to use in planning and governance. In each of these digital twinning concepts is a concept of what a city is. I noticed, looking at all of these, that there’s no people anywhere in any of these concepts.”

A research field far removed from Lotfi-Jam interests could potentially add guidance in populating digital twins for cities. Tianyi Peng, assistant professor in the decision, risk and operations division at Columbia Business School, is looking at the use of AI for decision making. His research looks at what can be used to generate AI agents that mimic human behaviour, such as that in the context of market decisions like shopping preferences and reactions to product stimuli.

The current use of digital twins for urban environments is limited for the lack of realistic representations of humans and their actions and interactions. It is easy to see how the study of individual behaviour and the simulation of group interactions will find use in city digital twins over time.

Peng’s colleague Olivier Toubia, professor of business at Columbia Business School, who is investigating interactions between AI-generated behaviour and how these interactions affect collective behaviour, “combines methods from social sciences and data science to study human processes such as motivation, choice, and creativity”.

Meanwhile, Lydia Chilton, associate professor of computer science at Columbia University School of Engineering & Applied Sciences, is contributing research into how AI agents in simulated environments can mimic unique behaviours of human interactions that can be unpredictable.

Providing comprehensive data

Mutualistic technologies offer ways looks at the wider set of technologies that interact with each other with impact on digital twins and robotics. The emerging network of mutualistic technologies features AI and sensors as the glue that creates positive feedback loops between these technologies. Data is needed to create realistic representations and relevant interactions between virtual and real world. Many times, real-world data can be difficult or expensive to extract though. Then synthetic data can find use. Synesthetic data can come from simulations in digital twins or from AI-based applications.

Commercial relevance of capturing comprehensive data is substantial, particularly for digital twins for urban environments where data from many interdependent networks require inclusion to realistically mirror activities and interactions of systems in cities. Road networks affect traffic patterns and public transportation impacts how people move through cities and therefore where businesses spring up.

Electricity networks, gas distribution, water and sewage lines crisscross urban maps and affect what neighbourhoods might lose power first during outages or which areas are prone to flooding. And flooding can affect power outages, which then can affect public transportation’s reliability, and so on. A very comprehensive view of urban activities is required to visualise interdependencies.

One of the general hurdles to effective and efficient city management are the silos in which urban networks and services operate. Data cannot easily connect; platform and format issues prevent seamless interfacing between systems, thereby posing genuine hurdles to all-encompassing digital twins that can truly capture and reflect the operational, commercial and social ongoings within cities. Therefore, a first step to creating genuinely beneficial urban digital twins often is a rather mundane, administrative step. Collaboration between administrations and agencies is key and the need for compatible data is crucial.

The city of San Diego’s managers realised the importance of such collaboration and created a partnership between the San Diego Association of Governments; the San Diego Regional Economic Development Corporation; San Diego State University; University of California, San Diego; and industrial partners. Interconnected dynamics and challenges in urban environments require connected, relatable data and digital twins that can represent resulting complexities – the collaboration of city administrations and network users is the first step.

Cautioning against developing communal divides

Digital twins will transform the way we plan, design, operate and maintain equipment, networks and urban environments. AI will accelerate their development, improve their performance and enhance their usability. But on the road to ubiquitous use, hurdles and issues need overcoming – some considerations generally associated with virtualisation technologies and use, others relate to AI, which currently is experiencing almost unchecked excitement and investment.

The Brookings Institution recently highlighted the emerging industrial and geographic unevenness of implementing and leveraging AI. The diffusion of AI will occur on different timelines in various industrial sectors. Spending on AI will depend on productivity and economic growth that companies and industries will expect or experience. Available investment capital and shareholder agreement will also play a role.

While it is natural that technology-related companies and finance, logistics and manufacturers firms already foresee substantial changes to their operations, agriculture, mining, personal services (including some aspects of healthcare) and many government services will see less immediate application opportunities. AI’s use for digital twins of cities will therefore initially create uneven representation in various sectors of urban planning and management.

Existing disparities between countries and regions will create geographical unevenness in the use of AI, and therefore in the adoption and diffusion of AI-empowered digital twins. Research by the Brookings Institution states: “Artificial intelligence is transforming the US economy, yet regional disparities in talent development, research capacity and enterprise adoption are stark and not yet fully understood.”

The digital divide emerged as a major concern at the end of the 1990s. Although the effects did not pan out as dramatic as some observers initially warned, the Covid pandemic from 2020 and following years highlighted unevenness in the way individuals, regions and entire countries were able to move personal and commercial activities online. Geographical laggards could develop in which AI implementation is slow, leaving other regions to charge ahead.

The report continues: “Such gaps and deficits may result in unrealised opportunities for productivity growth across disparate industries, and limit discovery and dissemination of the full range of AI use cases. For that matter, disparities in AI readiness may leave some communities to fall behind or slump into ‘development traps’. Imbalances in AI talent, innovation infrastructure and business adoption very well could decide which people and places will prosper in the future – and which will not.”

Internationally, such gaps can lead to “geo-algorithmic inequality”. Digital twins that replicate commercial activities, urban environments, entire ecosystems and eventually even economies as a whole will support the development of climate-resilience strategies, affect investment flows and establish the foundation for regional development plans in developed countries and metropolises.

“By contrast, much of Africa, the Caribbean and parts of Southeast Asia remain invisible in major digital twin ecosystems,” says the report. Data availability is spotty, often non-existent. Therefore, “decisions around infrastructure aid, disaster prevention or carbon offsetting are made with incomplete information – or without them in mind at all”.

Thinking globally, acting locally

The impact on these regions can be dramatic. Geo-algorithmic inequality results in “the uneven inclusion of countries and communities in the simulations that shape global policy, investment, and resilience planning”, according to Brookings Institution.

The effect can cascade towards digital twins that attempt to simulate the global ecosystem. In such virtualisations of the entire planet, structural bias can encode misrepresentations in digital twins and therefore distort resulting applications.

Potential approaches to alleviate such concerns exist. The Gaia-X initiative is looking to establish digital sovereignty by establishing “an ecosystem, whereby data is shared and made available in a trustworthy environment”. And the World Avatar effort is working on an “ecosystem of tools and services that can be used to create an individual digital twin, or network of connected digital twins, to provide a platform of data and model interoperability”.

Data silos of networks or country initiatives can then easily connect to each other. Although laudable, a concerted policy framework is needed to create incentives for corporations and organisations to buy into and fully embrace such efforts.

Martin Schwirn is the author of Small data, big disruptions: How to spot signals of change and manage uncertainty (ISBN 9781632651921) on foresight and horizon scanning. He is a strategy and innovation consultant for Global 2000 companies.

Source

Popular large language models (LLMs) are unable to provide reliable information about key public services such as health, taxes and benefits, the Open Data Institute (ODI) has found.

Drawing on more than 22,000 LLM prompts designed to reflect the kind of questions people would ask artificial intelligence (AI)-powered chatbots, such as, “How do I apply for universal credit?”, the data raises concerns about whether chatbots can be trusted to give accurate information about government services.

The publication of the research follows the UK government’s announcement of partnerships with Meta and Anthropic at the end of January 2026 to develop AI-powered assistants for navigating public services.

“If language models are to be used safely in citizen-facing services, we need to understand where the technology can be trusted and where it cannot,” said Elena Simperl, the ODI’s director of research.

Responses from models – including Anthropic’s Claude-4.5-Haiku, Google’s Gemini-3-Flash and OpenAI’s ChatGPT-4o – were compared directly with official government sources. 

The results showed many correct answers, but also a significant variation in quality, particularly for specialised or less-common queries.

They also showed that chatbots rarely admitted when they didn’t know the answer to a question, and attempted to answer every query even when its responses were incomplete or wrong. 

Burying key facts

Chatbots also often provided lengthy responses that buried key facts or extended beyond the information available on government websites, increasing the risk of inaccuracy.

Meta’s Llama 3.1 8B stated that a court order is essential to add an ex-partner’s name to a child’s birth certificate. If followed, this advice would lead to unnecessary stress and financial cost. 

ChatGPT-OSS-20B incorrectly advised that a person caring for a child whose parents have died is only eligible for Guardian’s Allowance if they are the guardian of a child who has died. 

It also incorrectly stated that the applicant was ineligible if they received other benefits for the child. 

Simperl said that for citizens, the research highlights the importance of AI literacy, while for those designing public services, “it suggests caution in rushing towards large or expensive models, which emphasise the need for vendor lock-in, given how quickly the technology is developing. We also need more independent benchmarks, more public testing, and more research into how to make these systems produce precise and reliable answers.”

The second International AI safety report, published on 3 February, made similar findings regarding the reliability of AI-powered systems. Noting that while there have been improvements in recalling factual information since the 2025 safety report, “even leading models continue to give confident but incorrect answers at significant rates”.

Following incorrect advice

It also found highlighted users’ propensity to follow incorrect advice from automated systems generally, including chatbots, “because they overlook cues signalling errors or because they perceive the automation system as superior to their own judgement”.

The ODI’s research also challenges the idea that larger, more resource-intensive models are always a better fit for the public sector, with smaller models delivering comparable results at a lower cost than large, closed-source models such as ChatGPT in many cases.

Simperl warns governments should avoid locking themselves into long-term contracts when models temporarily outperform one another on price or benchmarks.

Commenting on the ODI’s research during a launch event, Andrew Dudfield, head of AI at Full Fact, highlighted that because the government’s position is pro-innovation, regulation is currently framed around principles rather than detailed rules.

“The UK may be adopting AI faster than it is learning how to use it, particularly when it comes to accountability,” he said.

Trustworthiness 

Dudfield noted that what makes this work compelling is that it focuses on real user needs, but that trustworthiness needs to be evaluated from the perspective of the person relying on the information, not from the perspective of demonstrating technical capability.

“The real risk is not only hallucination, but the extent to which people trust plausible-sounding responses,” she said.

Asked at the same event if the government should be building its own systems or relying on commercial tools, Richard Pope, researcher at the Bennett School of Public Policy, said the government needs “to be cautious about dependency and sovereignty”.

“AI projects should start small, grow gradually and share what they are learning,” he said, adding that public sector projects should prioritise learning and openness rather than rapid expansion.

Simperl highlighted that AI creates the potential to tailor information for different languages or levels of understanding, but that those opportunities “need to be shaped rather than left to develop without guidance”.

With new AI models launching every week, a January 2026 Gartner study found that the increasingly large volume of unverified and low-quality data generated by AI systems was a clear and present threat to the reliability of LLMs.

Large language models are trained on scraped data from the web, books, research papers and code repositories. While many of these sources already contain AI-generated data, at the current rate of expansion, they may all be populated with it. 

Highlighting how future LLMs will be trained more and more with outputs from current ones as the volume of AI-generated data grows, Gartner said there is a risk of models collapsing entirely under the accumulated weight of their own hallucinations and inaccurate realities. 

Managing vice-president Wan Fui Chan said that organisations could no longer implicitly trust data, or assume it was even generated by a human.

Chan added that as AI-generated data becomes more prevalent, regulatory requirements for verifying “AI-free” data will intensify in many regions.

Source