After Verizon’s unexpected nationwide outage Wednesday, the carrier came out yesterday to try to save face, saying it was offering a $20 credit to all accounts in the myVerizon app. Now, its low-cost offshoot carrier, Visible, is also trying to make its own amends.
As spotted by 9to5Google, Visible is texting its customers with the following message, promising a $5 credit on their next bill as an apology for the outage:
Yesterday we let you down and for that we are sorry. We are giving you a $5 credit towards your next month of service that can be redeemed after Jan 16 when paying with a credit card online. This credit isn’t meant to make up for what happened. But it’s a way of acknowledging your time and showing that this matters to us. If you’re still having trouble connecting, please restart your device (power down and power back on).
The amount may be smaller, but it does seem to be a bit more convenient to claim, since it sounds like you can apply it while paying your bill (although I can’t make any promises for people using auto pay). Like Verizon’s offer, it also doesn’t seem to be discriminating between those affected by the outage and those who weren’t, as noted by a user on Reddit.
Some comments from other Visible subscribers on Reddit did point out that they had yet to receive the text, although I’m assuming that’s due to a slow rollout more than anything else, as textless users who reached out to customer service were able to confirm they were offered the $5 credit as well. I’ve reached out to Visible to confirm if the $5 credit applies to everyone, and will update this post when I hear back (I also asked if it would be applied to automatic payments).
As for why the payment is so much smaller, it’s probably due to the difference in cost between Verizon and Visible. Verizon said it based its $20 payout on the price for “multiple days of service,” and frankly, those same days cost much less on Visible.
While Visible uses the same network as Verizon, it sets itself apart by giving its users lower priority while connecting, which lets it offer unlimited plans starting at $20/month. Prices range to $33/month, but essentially, $5 goes a lot further on the low cost network, meaning it represents a similar value to $20 for a proper Verizon plan.
Whether that’s a fair price for an unexpected day of spotty connectivity, I’ll leave up to you.
A researchers’ propensity for risky projects is passed down to their doctoral students — and stays with trainees after they leave the laboratory, according to an analysis of thousands of current and former PhD students and their mentors. From a report: Science involves taking risks, and some of the most impactful discoveries require taking big bets. However, scientists and policymakers have raised concerns that the current academic system’s emphasis on short-term outcomes encourages researchers to play it safe. Studies have shown, for example, that risky research is less likely to be funded. Anders Brostrom, an economist studying science policy at the University of Gothenburg in Sweden, and his colleagues decided to examine the role of doctoral education in shaping risk-related behaviour — an area that Brostrom says has been largely overlooked.
“We often focus on thinking about how we can change the funding systems to make it more likely for people to take risks, but that’s not the only lever we have,” says Chiara Franzoni, an economist at the Polytechnic University of Milan in Italy. This study is “refreshing” because “we’ve discussed policy interventions a lot, but we haven’t discussed training,” she adds. […] The team found that students’ risk-taking dispositions matched those of their supervisors. This link was stronger when students and their supervisors communicated frequently, and weaker when students were also mentored by scientists outside their lab.
CyberGhost is the middle child of the Kape Technologies VPN portfolio, but in quality, it’s much closer to ExpressVPN than Private Internet Access. I mainly put it on my best VPN list because it’s so cheap, but I wouldn’t have done that if it didn’t earn its place in other ways — affordable crap is still crap, after all.
My universal impression of CyberGhost is a VPN that’s not perfect but is always genuinely working to make itself better. It makes decisions based on what will help its users, not to set itself apart in a crowded market. This makes it similar to a lot of other VPNs, but that’s not a bad thing — especially at such a low price.
Other than its price, the best things about CyberGhost are its intuitive app design, its frictionless user experience and the super-low latencies that make it an ideal pick for gamers. Download speeds are great up close but middling far away. While I love how many servers it’s got in Africa and South America, a few too many of them are virtual locations. I’ll get into all this and more in the review; feel free to read straight through or use the contents table to find the area that interests you most.
Editor’s note (1/16/26): We’ve overhauled our VPN coverage to provide more detailed, actionable buying advice. Going forward, we’ll continue to update both our best VPN list and individual reviews (like this one) as circumstances change. Most recently, we added official scores to all of our VPN reviews.Check out how we test VPNs to learn more about the new standards we’re using.
Findings at a glance
Category
Notes
Installation and UI
Windows app has more options and the most sensible organization
macOS app is very easy to use, but a bit lacking compared to Windows
Android and iOS both have simple main pages and slightly confusing preferences
No browser extensions (free proxy doesn’t count)
Speed
Excellent latency tests, with ping times short enough to lead the VPN field
Great download and upload speeds on close-in servers
Distant servers lag somewhat on both upload and download, bringing down the worldwide average
Security
Uses WireGuard, IKEv2 and OpenVPN protocols, but they aren’t all supported on all platforms
Blocks IPv6 and prevents WebRTC and DNS leaks
Disconnects when switching servers
Pricing
$12.99 per month
$41.94 for 6 months ($6.99 per month)
$56.94 for 28 months ($2.03 per month), renewing at the same price for 12 months ($4.79 per month)
Seven simultaneous connections
Bundles
Dedicated IP address for $2.50 per month
Privacy policy
Anonymizes all personal data
Can share data with other Kape subsidiaries, but only if they’re based in areas with good privacy laws
RAM-only servers and full-disk encryption confirmed by audit
Has never given information to police
Virtual location change
Unblocked Netflix perfectly in five different regions using streaming-optimized servers and WireGuard
Server network
125 server locations in 100 countries
Good global distribution, with nine locations in South America and six in Africa
However, most servers in the southern hemisphere are virtual locations that may not give the best speeds
Features
Kill switch cannot be turned off except on Windows
Split tunneling by app on Android and by URL on Windows
Content blocker can only be turned on or off, no customization
Large network of torrent-optimized servers and streaming servers
Smart Rules automation is both user-friendly and deep
Customer support
Online help pages are well-written but poorly organized
Live chat responds quickly; there is a bot but it’s easy to get past
Can also submit email tickets through an online portal
Background check
Founded in 2011 and based in Romania
Acquired by Kape Technologies in 2017
Installing, configuring and using CyberGhost
CyberGhost gets installation and UI largely right. There are no needless hurdles in the setup process. All its app designs put the important controls front and center and don’t overload you with needless information. I’ve broken up my thoughts by platform, as CyberGhost is pretty different depending on where you use it.
Windows
CyberGhost downloads and installs amazingly fast on Windows 11. I didn’t even have to grant any permissions. I just opened the .exe, clicked through a licensing agreement and logged into the desktop client. It took about two minutes end-to-end, including time I spent digging around in my password vault.
The CyberGhost VPN client for Windows 11.
Sam Chapman for Engadget
Once inside, you’re greeted with a UI that looks an awful lot like Surfshark — and if it’s not broken, don’t fix it. Options for special servers are on the left, the server list is in the middle and the connection interface is on the right. The arrangement prioritizes the most important controls and keeps clutter to a minimum. The only thing I can find to complain about is that clicking on a country with multiple locations doesn’t open the menu to choose between them; instead, you have to click on a hard-to-see arrow to the right of the name.
To access any of the special servers, click the appropriate tab in the left window, then choose from the list. Everything connects quickly. A gear icon at the bottom-left leads to all the special features and options, organized into three tabs: General (to do with the VPN app itself), CyberGhost VPN (to do with the VPN connection) and Account (to do with your subscription). The names could be better, but I can’t argue with the clear and useful descriptions on each feature.
Mac
The download process on macOS Sequoia is as easy as it is on Windows 11. CyberGhost walks you through every step, installs its helper tools with minimal fuss and is ready to go out of the box. It’s best to download directly from cyberghostvpn.com, since the App Store version is designed for iPads, not desktop computers.
Right after launching, the Mac app is pinned to the taskbar. To open it as a separate window, click the arrow button at the top-left. While it’s in the taskbar, the only things you can do are connect, disconnect and switch to one of your favorite locations. You can do all that from the standalone window, too, so there’s not much reason to ever leave it pinned.
CyberGhost’s VPN client for macOS Sequoia.
Sam Chapman for Engadget
The interface on Mac differs from the Windows client in other noticeable ways. For one thing, it’s permanently in dark mode, whereas Windows users get to choose between light and dark. There are fewer categories of servers in the left-hand column, with only torrenting (called For Downloading) and streaming options available — you can still connect to NoSpy, but only by going to the Romania location on the main list.
Also, the control panel gear is in the exact opposite location, sitting at the top-right of the connection window. The organization of options is also completely different and generally not as useful, with all the actual controls crammed into a single General tab.
This happened to me once or twice when my internet connection had no problems.
Sam Chapman for Engadget
Compounding the sense that CyberGhost didn’t give its Mac app as much attention as its Windows app, I kept getting the odd error pictured in the screenshot above. The client would tell me I had no internet connection (my internet was fine) and direct me to run a connection test. This would always turn up all green lights and let me connect without any incident. It rarely tripped me up for more than a moment, but it was still bizarre.
Android
CyberGhost on Android is streamlined to the extreme, focusing on ease of use above all else. Connections happen quickly, and the server selection is narrowed down, with only the streaming locations getting their own tab. It’s nice, but it does sometimes remind me of when I’d clean my room by shoving all the clutter under the bed.
CyberGhost connected on an Android phone.
Sam Chapman for Engadget
That’s mainly because tapping the gear at the top-right opens up a preferences menu with a lot more going on than the main screen. Most of the options here aren’t too complicated, but the shift is still jarring, especially since Android doesn’t do as well as Windows at explaining what everything does. “Anonymous statistics” and “Share network data for troubleshooting” sound like the same thing to me, and we get nothing on the esoteric concept of Domain Fronting.
Still, I’m nitpicking a bit. CyberGhost’s Android client does 95 percent of its job very well. Most of the settings aren’t necessary anyway, so you can pick your favorite server and be on your way.
iOS
Much like its Android app, CyberGhost’s iOS offering is sleek on the front end, a little cluttered in the back, but overall quite easy to use. Connections happen within seconds. The main page includes a useful option to tap on your current Wi-Fi network and immediately set Smart Rules for it. As with Android, only streaming-optimized servers and favorites are separated from the rest.
The main page of CyberGhost’s iPhone app.
Sam Chapman for Engadget
The control panel also looks very similar to what you get on mobile. The apparent clutter comes from simple on-off toggles and more complex submenus being all jumbled up together, but you can use the VPN just fine without engaging with any of it. For the most part, CyberGhost on iOS does a lot to help you and nothing to get in your way.
Browser extensions
CyberGhost doesn’t have a full-service browser extension. If you look for an extension link on the download hub, you won’t find anything. What it does have is free proxy add-ons for Chrome and Firefox, which can be used to change your IP address to a new location.
However, proxies do not encrypt your traffic, leaving out one of the critical aspects of how a VPN works. The extension library pages for the CyberGhost proxies are vague about this, but they’re no substitute for a full VPN. They’re free and may be convenient for occasional streaming if they don’t get caught, but they aren’t secure.
CyberGhost speed test
I conducted all these tests on a wireless connection using the WireGuard protocol. For each, I selected either a physical server or a virtual location close to its physical source. Here’s what each metric means in the table below:
Ping, measured in milliseconds (ms), is a measure of latency — how long it takes to send a signal from your device to its destination via the VPN server. Lower pings are better. Since signals can only move so quickly, latency tends to increase with distance.
Download speed, measured in megabits per second (Mbps), is what you probably think of as “internet speed” — how fast websites load and how much video you can stream without any pause to load.
Upload speed, also measured in Mbps, determines the rate at which data travels from your device to its destination. It’s useful for posting content, saving files to the cloud, torrenting and two-way video calls.
Server location
Ping (ms)
Increase factor
Download speed (Mbps)
Percentage drop
Upload speed (Mbps)
Percentage drop
Portland, USA (unprotected)
16
—
58.70
—
5.80
—
Seattle, USA (fastest location)
22
1.4x
55.88
4.8
5.60
3.4
New York, USA
155
9.7x
45.43
22.6
5.43
6.4
Montevideo, Uruguay
111
6.9x
46.25
21.2
5.55
4.3
Lisbon, Portugal
328
20.5x
45.60
22.3
4.36
24.8
Johannesburg, South Africa
632
39.5x
34.12
41.9
3.68
36.6
Vientiane, Laos
350
21.9x
38.04
35.2
4.78
17.6
Average
266
16.6x
44.22
24.7
4.90
15.5
CyberGhost’s speed test gave me mixed results — mostly good, but with some reasons for caution. To start on the positive side, latency results were excellent. No matter where I went in the world, the numbers only jumped above 400 milliseconds in one place, and that was the Johannesburg server that had problems across the board. CyberGhost’s 266 average is significantly better than I got when testing Surfshark, currently the fastest VPN overall.
A speed test using the fastest location chosen by the CyberGhost app.
Sam Chapman for Engadget
Download and upload speeds looked good in my fastest location, Seattle. Using CyberGhost only slowed my browsing by 4.8 percent and dropped my upload rates by 3.4 percent, comparable to most of its leading competitors. At a distance, though, speeds started to falter. Things in New York remained reasonably fast, but with a lot of fluctuation between tests; unlike Seattle, numbers swung between the 30s and 50s.
As I virtually traveled the world, I saw more and more swings, plus sharp declines in South Africa (which is always the problem child of VPN servers, for some reason). To put this in perspective, CyberGhost never dragged that much on my browsing speed, and the internet remained usable no matter where I went. It’s just slightly more sluggish than my favorite VPNs in every area — except latency, where it soars ahead.
CyberGhost security test
VPNs need to secure your internet activity against two things: intentional attacks and leaks due to negligence. A VPN should be watertight enough that it never lets your information slip by accident, while also defending your data against outside interference.
It’s not hard to test whether a provider is meeting these two criteria. First, make sure it’s using safe VPN protocols with modern encryption. Second, use an IP address checker to test for DNS, WebRTC and IPv6 leaks. Third, test encryption itself to ensure it’s being applied equally to all data packets. Let’s get started.
VPN protocols
CyberGhost supports three different VPN protocols, all of them up-to-date and secure. OpenVPN, available on Windows, Android, Linux and Fire TV, is my typical recommendation, balanced and secured through a multi-decade history of refinement. WireGuard, supported on every platform, is the new hotshot on the block, fast and stable but not quite as rigorously tested as OpenVPN. IKEv2, which works on macOS, iOS and Windows, connects more quickly than the others but isn’t open-source.
I have some quibbles about how available these protocols are. OpenVPN should always be an option for everybody; leaving it off Apple devices doesn’t make sense. I’ve asked CyberGhost about this and will update here when I get a reply. In the meantime, I can’t complain about the protocols themselves, which use uncracked encryption ciphers and present no obvious weak points.
Leak test
I used ipleak.net to check CyberGhost for leaks. There are three likely causes for a VPN to accidentally reveal your real IP address: it failed to account for different IP types (IPv6 leak), a real-time connection went outside the encrypted tunnel (WebRTC leak) or it used a domain name server that an ISP could read (DNS leak).
CyberGhost never leaked my real IP address.
Sam Chapman for Engadget
CyberGhost blocks all IPv6 traffic, so there’s no chance of an IPv6 leak. I checked for WebRTC leaks and didn’t find any. Likewise, every time I connected to a VPN server and refreshed the page, I saw that server’s IP address, proving that CyberGhost isn’t leaking.
There is one important exception here: whenever you select a different server on your CyberGhost client, it disconnects from that server before connecting to the next one. This means that your data is exposed during the transition. It’s annoying, but as long as you remember not to do anything risky while changing locations, you’ll be fine.
Encryption test
For my final test, I used WireShark to capture images of the data packets CyberGhost was routing from my device to its servers. Sure enough, the outer layer of each data stream was encrypted no matter which VPN protocol I used. Ultimately, all my probing showed that CyberGhost is secure against both negligence and interference.
How much does CyberGhost cost?
CyberGhost sells three different subscriptions, all of them with the same features. The only difference is how long the plan lasts. You can save money overall by signing up for six months or two years at a time, but it costs more upfront. Each plan can be used on seven devices simultaneously.
These prices are subject to change.
Sam Chapman for Engadget
One month of CyberGhost costs $12.99, and it renews automatically at the same price at the end of each billing month. Each monthly renewal comes with a 14-day, money-back guarantee. You can get six months for $41.94 total, which works out to $6.99 per month. At the end of six months, it auto-renews at the same price. The six-month plan comes with a 45-day refund guarantee.
Finally, there’s the two-year plan, which comes with a lot of fine print. The first time you sign up, it costs $56.94, which gets you a total of 28 months (working out to $2.03 per month). However, after your 28 months are up, all subsequent renewals instead get you a 12-month plan — still for $56.94, but now working out to $4.79 per month. That’s still relatively cheap, but not nearly as affordable as some VPNs with perpetual two-year plans.
CyberGhost side apps and bundles
CyberGhost doesn’t have much in the way of additional subscription offers, but that’s honestly refreshing. In an age when even the best providers also need to be antiviruses, insurance agents and probably vacuum cleaners, it’s nice to see a VPN that’s content with just being a VPN.
CyberGhost does have a broader “security suite,” but it comes at no extra cost and is currently available on Windows only. More info on that in the “Extra features” section below.
Dedicated IP
You can pay an extra $2.50 per month to add a dedicated IP address to your CyberGhost plan. With a dedicated IP, you’ll have a stable address whenever you get online through the VPN, which makes it a lot easier to connect to firewall-protected web services. It’s also exclusive, so nobody else can get you in trouble by misusing the IP address.
Close-reading CyberGhost’s privacy policy
CyberGhost is located in Romania, which makes it subject to the strict privacy laws of the European Union. It’s not legally required to keep tabs on its users or install government backdoors. That’s a great start, but to be certain about a company’s approach to privacy, it’s best to look at its own words.
A VPN makes your online activity anonymous to anyone else who tries to look at it, but the VPN itself still has the power to violate your privacy if it chooses. This leads some people to advise against using commercial VPN services at all, though I don’t go that far. The best VPNs build in features that make it impossible for them to abuse their access to your web traffic, such as RAM-only servers and full-disk encryption.
When trying to determine if you can trust a VPN with your privacy, the first place to look is its official privacy policy. This document lays out everything the VPN does to handle your personally identifiable information (PII). If the provider violates its policy, they can be sued, so it’s not in their interest to lie outright in the document.
I went over CyberGhost’s privacy policy with a fine-toothed comb — it can be found here if you’d like to follow along. It starts with the usual promise of “uncompromising protection”: CyberGhost swears that “we are NOT storing connection logs, meaning that we DON’T have any logs tied to your IP address, connection timestamp or session duration” (all emphasis theirs).
That’s the standard I’ll be checking against: a total lack of any way for CyberGhost to read or share information on its own users. Let’s see how it holds up.
These may just be words, but they have legal force, at least in civil court.
Sam Chapman for Engadget
The privacy policy wins early points by clarifying all the data it collects. You can see the whole quote in the screenshot above, but to summarize, any PII (like your email or IP address) will never be connected with anything you do online. Since absolute anonymity is impossible, this is the best we can hope for from a VPN.
Later on, the policy clarifies everything CyberGhost might do with personal data, none of which involves turning it over to authorities or selling it to advertisers. The most suspicious reasons given are “fraud detection/prevention” and “To enforce the terms of service,” but these both relate to kicking users off CyberGhost itself, not tattling on them to the government.
The only potential problem comes in the section titled “Sharing Your Personal Data.” Here, CyberGhost states that “we may communicate your personal data to a member of our group of companies,” meaning all subsidiaries of Kape Technologies. I won’t rehash the case against Kape in full — my ExpressVPN review covers it in detail.
Suffice to say the only real risk here is that CyberGhost might share PII with another Kape company located in a region with worse privacy laws than Romania or the EU. To me, this isn’t a serious concern. First of all, Kape doesn’t own any companies based in truly anti-privacy nations like China, India or Russia.
Moreover, the privacy policy states that CyberGhost won’t share information with any entity not “located in the EU or another jurisdiction offering equivalent data protection standards.” Every bit of data gets the same protections. This may mean PII enters a country in the Five/Nine/14 Eyes alliance, but the Eyes only matter if a VPN is already logging data it shouldn’t have. It’s not that abuse of intelligence-sharing agreements isn’t a problem; it’s just that the risk it poses starts with the VPN itself, not where it’s located.
To sum up, I didn’t see any red flags or loopholes in the CyberGhost privacy policy. Some clauses could be tightened up, and it always pays to be suspicious, but I’m confident that using this VPN doesn’t risk your personal privacy.
Independent corroboration
CyberGhost has been audited twice by Deloitte Romania, once in 2022 and again in 2024. Following that pattern, I’ll be looking out for another one this year. You need an account to read the full audit report, but it’s only 10 pages and easy to summarize: the auditors found nothing in CyberGhost’s systems that conflicted with its privacy policy.
The audit notes CyberGhost’s server infrastructure as evidence. All servers are run on RAM with full-disk encryption, making any information they store completely ephemeral. Even if CyberGhost staff wanted to spy on you, they wouldn’t see anything. The same goes for third-party hackers.
CyberGhost also posts a regular transparency report that lists how often law enforcement has asked it for information. As far as I could find, after hundreds of requests, there’s never been a case where CyberGhost provided any information to cops.
Can CyberGhost change your virtual location?
For this section, I used Netflix to test whether CyberGhost’s virtual location changes are detectable by other websites. Ideally, every time I change location with CyberGhost, Netflix would accept it as real and show me the content library from that country. If either of those things doesn’t happen within three tests, the VPN has a problem.
Since CyberGhost has servers built for streaming, I used those for each of the five locations. You can see my results below.
Server location
Unblocked Netflix?
Changed content?
United Kingdom
3/3
3/3
Japan
3/3
3/3
Germany
3/3
3/3
Australia
3/3
3/3
Brazil
3/3
3/3
This test was a smashing success for CyberGhost. Every time, it showed me the proper video library for the location I chose and never once got caught by Netflix’s firewalls. It’s the best result I’ve seen in this section since I tested Proton VPN, and that’s high praise if you know me.
CyberGhost has 125 server locations in 100 countries. Of those locations, 75 are real and 50 are virtual, which makes the math easy: CyberGhost’s VPN server network is 60 percent bare-metal and 40 percent virtual. That’s good, since physical servers let you calculate how much performance will deteriorate over distance — virtual servers are just as safe, but speeds might fluctuate depending on where they really are.
Region
Countries with servers
Total server locations
Virtual server locations
North America
9
21
5
South America
9
9
9
Europe
45
56
13
Africa
6
6
3
Middle East
6
6
4
Asia
23
23
16
Oceania
2
4
0
Total
100
125
50
Looking at the distribution of servers, we get good news and bad news. The good news is that there really are 100 different countries and territories to choose from, encompassing nearly all the virtual globetrotting you’re likely to need. There are also lots of servers in the southern hemisphere, which is often the last place VPNs grow into. There’s a wealth of choices in South America, plus several options in Africa and Central Asia.
CyberGhost’s selection of VPN servers.
Sam Chapman for Engadget
The bad news is that the distribution of real servers is skewed toward Europe and the United States. None of the nine South American servers are actually located in South America; worse, a large number of them are physically located in Miami. If you’re using CyberGhost in Argentina, don’t expect top speeds from the Buenos Aires server, since it’s actually over 4,000 miles away. CyberGhost’s support center does include a list of where the virtual servers are relayed through, but it’s not up to date.
Extra features of CyberGhost
CyberGhost has a few features beyond the VPN itself, though not as many as you might think. Compared to a provider like NordVPN, which goes all in on extra features, CyberGhost’s offerings look pretty lean. But that doesn’t matter as much if the features work well. Let’s see how they do.
Kill switch
CyberGhost takes an unusual approach to its kill switch. In case you aren’t familiar with the term, a kill switch cuts off your internet connection if your link to the VPN ever drops, protecting your anonymity in case of unexpected incidents. Most VPNs let you toggle the kill switch on and off, but on CyberGhost, it’s fully engaged 100 percent of the time — except on Windows, where you can turn it on and off as desired.
Turning on the kill switch is almost always a good idea, but it’s still annoying that Cyberghost gives many of its users no way to turn it off. In rare cases, kill switches can get overzealous, preventing you from getting online even when conditions are safe. It’s an odd choice to remove a potential troubleshooting step from the user’s control.
Split tunneling
Split tunneling lets you name some apps or websites that will run unprotected even while the VPN is active. This can help with certain services that refuse to work if they detect a VPN, or alternatively, can protect only one sensitive app or site while the others enjoy faster unprotected speeds.
CyberGhost only has full split tunneling on Android. It also offers a slightly different feature called Exceptions on Windows. Android split tunneling works by app, while Exceptions works by URL. In both cases, you choose individual apps or websites to leave out of the VPN. It’s limited, but works as advertised.
Optimized servers
As I mentioned in the Netflix testing section, CyberGhost includes specialized servers designed for specific tasks. Other than the add-on dedicated IP servers, these come in four forms: “For gaming,” “For torrenting,” “For streaming” and “NoSpy.” Gaming servers are apparently built to keep latency low, but I couldn’t see much difference between them and the normal servers.
“For torrenting” is called “For downloading” on Mac, but it’s all the same torrent-optimized servers. These are built to meet the download and upload speed requirements for effectively using P2P filesharing clients. CyberGhost has P2P servers in 86 countries, which makes it a good VPN for torrenting; only the lack of port forwarding keeps it from being truly great.
A few of CyberGhost’s specialty servers on a MacBook.
Sam Chapman for Engadget
Each streaming server is built to unblock a specific streaming site in a particular country, occasionally for a single type of device. For example, United States streaming servers are aimed at Netflix, Hulu, Peacock, Disney Plus, Amazon Prime Video and more, many in their Android or Smart TV forms. UK servers work for Netflix UK, BBC iPlayer, ITV and more. In total, there are 106 streaming servers in 22 countries — not quite as extensive as the overall list, but it’s important to remember that non-optimized servers still work fine for streaming.
Finally, the NoSpy options connect to a set of servers in Romania that CyberGhost claims to manage entirely in-house, with nobody able to access them except CyberGhost’s own team. This is good, but it leaves me suspicious about who’s running the rest of the servers. Are they all run by third parties except the NoSpy locations? That’s relatively common, but it creates vulnerabilities if the VPN provider doesn’t insist on high standards from collaborators.
Content blocker
CyberGhost’s content blocker is underwhelming. All you can do is turn it on and off. There’s no customization like you get with Windscribe’s R.O.B.E.R.T. and no clear statement of where it’s getting its list of domains to block. In practice, it does block in-page ads, but without specifics I couldn’t test it in more detail.
There’s no customization on CyberGhost’s blocker — just turn it on or off.
Sam Chapman for Engadget
Smart Rules
The Smart Rules automation suite is the crown jewel of CyberGhost’s features and the most common reason I recommend it. Using Smart Rules, you can automate CyberGhost’s behavior to a degree inconceivable on most other VPNs.
You can program CyberGhost to take different actions on each of your usual networks.
Sam Chapman for Engadget
Smart Rules come in two forms: actions performed automatically when CyberGhost launches or connects and actions that respond to new Wi-Fi networks. In the former category, you can set CyberGhost to connect when you open the app, determine which location it connects to and even set an app to automatically open after it connects.
Wi-Fi rules depend on whether the network CyberGhost detects is secured or not. For each type of network, you can set the VPN to connect, disconnect, ask you what to do or ignore it entirely. Once it recognizes a Wi-Fi network, you can set specific rules for that network. It’s at once very easy to use and capable of surprising depth.
CyberGhost customer support options
CyberGhost primarily offers customer assistance through its online portal, which can be reached at support.cyberghostvpn.com or by going through the app. If you choose the options “CyberGhost VPN help” or “FAQ” in the app settings, you’ll be taken to the support pages in a browser. I recommend going through the URL, since that takes you to the highest-level page.
The support center feels distressingly like an afterthought. Written guides are divided into four sections: Guides, Troubleshooting, FAQs and Announcements. The latter has only one article and the former three are roughly interchangeable — if I’m having trouble connecting to a server, is that an FAQ, a Guide or Troubleshooting? Looking for any particular subject here is a needle-in-a-haystack search.
Fortunately, there is a search bar, but this presents its own problems. A simple search for “connection issues macos” turned up 72 results, including one called “Troubleshooting connectivity on macOS” and another titled “Troubleshooting VPN connection on mac.” These two articles are in different sections, but mostly contain the same information, except that the former has an extra walkthrough on renewing your DHCP lease.
It’s a shame, because the articles themselves are mostly clear and helpful, with lots of well-chosen screenshots. Someone clearly worked hard on the content, but the overall organization left me thinking the knowledgebase was thrown together years ago and hasn’t been checked since.
Live support experience
If you have trouble finding what you need in the written guides, you can get personal support in two ways. One option is to submit an email request through a Zendesk portal. This gives you all the time you need to frame your question and add supporting materials, at the cost of waiting longer for a reply.
Your other option is to access live chat, which you can do from anywhere on cyberghost.com by clicking the chat button in the bottom-left corner of the screen. Live chat starts with a “CyberGhost AI Assistant” (what we used to call a chatbot in the good old days) which runs you through several diagnostic questions. To its great credit, the bot does not try to link you to articles in the knowledgebase, understanding — as too many providers don’t — that nobody tries live chat unless the FAQ isn’t working for them.
It didn’t take me too long to get in touch with what was apparently a live expert.
Sam Chapman for Engadget
I decided to bother CyberGhost about the connection issue I wrote about in the Mac UI section. Within seconds, the chatbot offered me a button that would transfer me directly to a live agent. I only had to wait about 4 minutes before the agent got in touch. After that, each response took about a minute and explained everything carefully and efficiently. It was as helpful as the written knowledgebase wasn’t.
CyberGhost background check
CyberGhost was founded in 2011 in Bucharest, Romania, where it’s still headquartered today. It claims to have around 38 million subscribers and a staff of 70. It appears to be most popular in France and Germany.
The only thing that makes me at all uncertain about CyberGhost is that I can’t find much information about its history — it doesn’t even have a Wikipedia page in English. By far the most likely explanation is that CyberGhost is exactly what it seems to be: a reliable, drama-free VPN provider that doesn’t court controversy. Still, I’m naturally paranoid, so I’d understand if this lacuna sends you running back to a better-documented VPN.
There is precisely one date in CyberGhost’s history that everyone lists: 2017, when it was purchased by Kape Technologies. As a VPN reviewer, I have to think about Kape a lot. My opinion is that the fear around it doesn’t measure up to reality. For example, back when it was known as Crossrider, Kape was not a “malware distributor”; it sold an ad-injection plugin that turned out to be a useful malware vector.
Perhaps Crossrider could have worked harder to stop its platform from being misused, but that doesn’t make it a security threat today. Similarly, being owned by a businessman from Israel does not mean that Kape or CyberGhost are secretly controlled by Mossad.
I’m not here to defend Kape — I’m just pointing out that a lot of the fear isn’t backed up by evidence. To my mind, Kape’s consolidation of the VPN industry (it also owns ExpressVPN and Private Internet Access, plus two websites that review VPNs) is bad enough without having to look for additional conspiracies. It’s up to you to decide whether or not CyberGhost’s parent company presents a hard line you won’t cross.
Final verdict
At the end of my journey with CyberGhost, I may not be blown away, but I’m definitely pleased. After my poor experience with PIA, I was afraid the only budget VPN I could wholeheartedly recommend was a two-year subscription to Surfshark. CyberGhost is a meat-and-potatoes VPN — it’s not pushing any envelopes, but it’s cheap and it does the job.
All that said, I recommend it more to casual users than to people who really need secrecy. There are just enough reddish flags that I wouldn’t necessarily trust it with life-and-death information: the (possible) use of third-party managers for all servers outside Romania, the freedom to share information with any Kape subsidiary, the loss of encryption when switching servers. It’ll keep you anonymous and let you stream foreign TV for cheap, but you should still choose Proton VPN if you need serious privacy.
This article originally appeared on Engadget at https://www.engadget.com/cybersecurity/vpn/cyberghost-vpn-review-despite-its-flaws-the-value-is-hard-to-beat-200000250.html?src=rss
The launch of Android XR, a newly announced headset from Valve, and major strategic shifts at Meta, are just the start of what made 2025 a significant year for the XR industry at large. 2026 will mark the 15th year that we’ve been following this XR journey here at Road to VR. With the context that comes with that long-term perspective, it’s once again time to reflect on the biggest stories of the last year and to talk about what’s on the horizon.
Meta Makes Aggressive Cuts as It Shifts XR Strategy
“This year likely determines whether this entire [XR] effort will go down as the work of visionaries or a legendary misadventure,” Bosworth wrote.
Well, by early 2026 it’s looking like “legendary misadventure” takes the cake.
Apparently not seeing the growth and traction it wanted, Meta is making an aggressive shift in its XR strategy. The last few days have seen reports of multiple first-party VR studios being cut down to size or outright shuttered. The company’s business-focused virtual collaboration space, Horizon Workrooms, is being shut down, and more. In total, the company is said to be cutting roughly 10% of its entire Reality Labs division as it shifts focus away from VR and “metaverse” efforts.
What it Means for 2026
It’s going to take a while for the dust to settle on this, and it probably won’t be until Q2 that the company shares a clear vision for what it hopes to accomplish with the cuts and new direction.
From what I’m seeing and hearing, it sounds like Meta isn’t exiting the XR space, but it’s shifting focus more strongly toward the glasses end of the spectrum, while doing away with the notion that building a “metaverse” (a digital space where people would gather, play, and work) is a strategic imperative.
Rather, it looks like Meta will continue to run its XR headset platform and let it evolve naturally rather than trying to place big content bets or force the metaverse into existence. Meanwhile, the company is said to be focused on boosting production of its smart glasses to serve growing demand.
Although that likely means a greater focus on smart glasses and AI assistants for the time being, it’s clear that the company’s end-goal is (and has been) to evolve its smart glasses into full-blown augmented reality glasses over time. In fact, Meta showed an early vision of this end goal back in 2024 with the Orion prototype. The 2025 release of the Ray-Ban Display glasses, and the ‘neural band’, was a clear step toward that goal.
Ray-Ban Display is still just a pair of smart glasses (ie: a small field-of-view and a static display with no tracking). But already the company that makes the waveguide in Meta’s glasses says it has a much larger 70° field-of-view waveguide that’s ready for production.
For many years I’ve explained that the industry has been working on the challenge of compact and affordable XR devices from two sides. On one hand, the industry has started by packing a wishlist of features into a bulky headset, and then trying to make it smaller. On the other hand, the industry is starting a tiny glasses-like package, and then trying to add back all the features enjoyed by the bulkier headsets.
Meta has mostly focused on the former (headsets), but it’s now shifting focus to the latter (glasses). The end goal, however, remains the same: an affordable and comfortable device that can digitally alter the world around you.
This big shift isn’t just a big deal for Meta, it’s a big deal for the whole industry. Meta has been dominant in the space for years, thanks in a big part to being able to out-price the competition and attract developers, thus building the leading standalone headset platform. With its standing in the industry, Meta has been able to direct much of what has happened within the industry, either explicitly or implicitly.
If Meta is pulling back on its VR and metaverse initiatives, the door may open for another company to take over its influential role. Or, the space might settle into a new equilibrium with a renewed competitive landscape, which has long been suffocated by Meta.
All we can say for certain is that 2026 will be a year of major realignment as the industry figures out how hands-on Meta plans to be with its VR platform going forward.
The Biggest Year in Recent History for XR Hardware
2025 turned out to be a huge year for XR hardware launches and announcements.
Google finally revealed and launched Android XR, a direct competitor to Apple’s VisionOS. Samsung launched the first Android XR headset, Galaxy XR, a direct competitor to Apple’s Vision Pro.
Meanwhile, Valve is taking a whole new approach to its VR architecture. Frame is a fully standalone headset—a first for Valve—and the company has designed it to be a better companion to people’s existing Steam library, by allowing it to play pretty much any Steam game (VR or otherwise) either locally or streamed from a nearby PC.
While Valve is giving PC VR some much needed love, I’m still not convinced that Frame is going to revolutionize the space. Although it has some neat extras (like improved wireless game streaming thanks to eye-tracked optimizations), it doesn’t really do that much more than a Quest 3 or Quest 3S, which will inevitably be the cheaper options. As with its prior headset, Frame will probably remain limited to an enthusiastic niche of hardcore PC VR players. But ultimately, Frame shows that Valve never stopped caring about VR and that the company is still focused on making Steam an open VR platform on PC that will be maintained for years to come.
Meta has been hard at work prototyping full-blown AR glasses, but it hasn’t actually launched such a product yet. Meanwhile, Xreal and Viture have been rapidly evolving their smart glasses with growing AR capabilities, seemingly catching companies like Meta by surprise. The pair of $100 million investments into Xreal and Viture (and especially Xreal’s close partnership with Google) will put pressure on Meta to release its AR glasses sooner rather than later.
Valve Reveals a New Headset, But Confirms No New First-party VR Game to Go With it
Given that Valve launched Half-Life: Alyx back in 2020 to show what was possible with its first VR headset, there was widespread speculation that the company would similarly announce a new VR game to launch alongside Steam Frame. But as the company told me directly, there is no new first-party VR game in development.
What it Means for 2026
The lack of a flagship launch title to go out the door with Steam Frame has left many scratching their heads. New headsets are exciting, but given the dearth of exciting PC VR content in the last few years, what are people actually going to play… more Beat Saber?
I’m glad Valve is still in on VR, but I’m not exactly bullish on Frame. Luckily the PC VR landscape has never had more options to serve various hardcore PC VR niches, thanks to companies like Pimax, Bigscreen, and Shiftall—and hey, even Sony technically makes a PC VR headset!
A Shifting VR Player Demographic Comes to a Head as Veteran VR Studios Struggle to Stay Afloat
2025 was a brutal year for established VR studios. Highly immersive single-player apps were once the bread and butter of VR gaming. But VR was not insulated from the broader gaming industry shift toward free-to-play multiplayer games.
That shift seems to have reached a peak just as a wave of prior long-term bets on single-player VR content was coming to fruition in 2025. The result has been report after report of established VR studios struggling to stay afloat.
On the other hand, new studios focused on free-to-play multiplayer content have seen rapid growth and seemingly reached unprecedented new peaks of player counts and retention. Games like Gorilla Tag, Animal Company,Yeeps, and UG are dominating Quest’s Top Selling Charts by serving a younger demographic of players looking for free-to-play multiplayer experiences. Interestingly, all four of these newer top-earning titles are also built around arm-based locomotion.
What it Means for 2026
Whether we like it or not, free-to-play multiplayer is here to stay. Many of the most popular non-VR games are free-to-play multiplayer games, so it should be no surprise that the same formula would take over VR as well. The unfortunate part is that the transition happened so fast that by the time the latest wave of big budget single-player VR games landed, they were launched into a void of demand. With production times of some bigger VR games spanning 1-3 years, it’s difficult to course-correct.
Especially with Meta’s latest cluster of studio closures, the message is now unambiguous: premium single-player VR games are no longer what the bulk of active VR users are looking for. That’s not to say there’s no room for great single-player experiences in VR, but the demand for them isn’t what it used to be.
Of the veteran VR studios that have managed to weather the storm, I expect to see many of them take their first stab at free-to-play multiplayer VR games, or focus on ‘VR optional’ titles, or even leave VR for the time being while they seek greater stability in the larger gaming market.
Frankly, I think this situation has a bit less to do with the ‘free-to-play’ part, and more to do with the ‘multiplayer’ part. As with almost every entertainment activity in existence, most people like to play games with their friends. The rise of massively successful paid multiplayer games with structures that are reminiscent of traditional single-player games (ie: Destiny, Valheim, Helldivers, Arc Raiders) tells me that pure single-player games as a whole will one day become a thing of the past.
That’s not to say that we won’t see great, ‘single-player style’ games still made (like, say, Red Dead Redemption 3) but I bet you’ll at least have the option to play them with a friend or two.
These changes were all clearly meant to address first-generation pain points. Specifically, the improved headstrap was a major admission that the headset was too heavy and bulky. Unfortunately a better head strap can only do so much.
I don’t expect we’ll see any new XR hardware from Apple in 2026. But I do expect to see the company continue to make more of these first-generation fixes and to further improve the headset’s most promising use-cases on the software side. I’m still personally hoping for better window management.
While there’s been much reporting about Vision Pro as a ‘failed’ product, those that are actually connected to the XR industry understand that Vision Pro is a significant contribution to the state-of-the-art that’s really only held back by its current size and weight. I’m certain Apple knows this too.
My bet is that Apple is far from done with Vision Pro and VisionOS. It’s rare for the company to make a product play only to cancel it after one generation. More likely, I’m willing to bet that Apple has set new and specific goals for the size and weight of its next Vision headset, and will happily wait for years until it can actually meet those goals. In the meantime, it will continue to invest in VisionOS, which I’ve long said is a more important contribution to the industry than the headset hardware itself.
First released in 1995, the Virtual Boy was portrayed as a type of “virtual reality” experience, but considering its small field-of-view, lack of motion tracking, and single-color (red) display, it was functionally just a 3D display on a stand. Still, the console has been culturally associated with “virtual reality” ever since—and it’s not exactly a positive association.
Ambitious as it was, Virtual Boy was an infamous failure of a game console, owed largely to its minimal game catalog, single-color display, and reports of motion sickness while playing. It was discontinued less than a year after launch.
The upcoming $100 accessory will use Switch or Switch 2 as the brains (and display) of the device, and it will play original Virtual Boy games like Mario’s Tennis, Teleroboxer, and Galactic Pinball, with a planned total of 14 titles to be launched in time (that may not sound like many, but it’s more than 50% of the entire Virtual Boy game catalog).
Nintendo will also sell a $25 ‘cardboard’ version of the Virtual Boy accessory which will allow Switch to play the same games but without the stand and plastic facade to hold the console.
We still don’t know if the games are simply being emulated or if they have been retouched or remastered. I hope they’ll be at least updated to render at the native Switch or Switch 2 resolutions, rather than the tiny 0.086MP (384 × 224) per-eye resolution of the original Virtual Boy.
In my book, the biggest wild cards for 2026 are Snap and HTC.
HTC was once a prominent player in the VR space, having built a long line of PC VR headsets that rivaled Meta’s Rift. But once Meta shifted focus to standalone, HTC wasn’t able to keep up. Sure, HTC released several standalone headsets, but none have come close to the consumer and developer traction of Meta’s Horizon.
Exactly where HTC heads next is unclear. Will it follow Meta’s lead again and shift its primary focus to smart glasses? Or could it swoop in and try to fill the vacuum left by Meta’s pullback from the VR and metaverse space?
The latter could be a significant opportunity for the company which, at very least, has the same core pieces already in place (standalone VR headsets, an app store, and a ‘metaverse’ platform). Not to mention strong traction in the B2B and LBE spaces, which Meta never quite got a handle on.
The company has launched two generations of its ‘Spectacles’ AR glasses, and the company has spent time focusing on developers and building out tooling based on feedback.
Snap plans to launch its first pair of consumer AR glasses this year, but it remains to be seen if it has any unique technological advantages compared to what’s already out there. Even if not, it’s possible that Snap’s social and fun-focused approach to AR glasses could be a winning play, especially if it can successfully draw its fleet of Snapchat AR developers over to its glasses. The company says that’s the plan, anyway, as it has been building tools that make it easier for developers to build Lenses that span both hand-held and head-worn AR.
– – — – –
As someone who has been reporting on this industry for nearly 15 years now, I truly mean it when I say I believe 2025 will be looked back upon as one of the most significant moments for the XR industry overall. The next five years are certain to see more change, competition, and innovation than the last five years.
Big Tech’s AI-fueled memory shortage is set to be the PC industry’s defining story for 2026 and beyond. Standalone, direct-to-consumer RAM kits were some of the first products to feel the bite, with prices spiking by 300 or 400 percent by the end of 2025; prices for SSDs had also increased noticeably, albeit more modestly.
The rest of 2026 is going to be all about where, how, and to what extent those price spikes flow downstream into computers, phones, and other components that use RAM and NAND chips—areas where the existing supply of products and longer-term supply contracts negotiated by big companies have helped keep prices from surging too noticeably so far.
This week, we’re seeing signs that the RAM crunch is starting to affect the GPU market—Asus made some waves when it inadvertently announced that it was discontinuing its GeForce RTX 5070 Ti.
Can you smell what ASRock is cooking? Tired wrestling references aside, what ASRock is cooking looks pretty darn tasty — especially for something called “Rock”. Four new motherbords in a new family are the inaugural components: the Socket AM5-based B850 Rock Wi-Fi and the LGA 1851-equipped B860 Rock Wi-Fi, both available in full-sized ATX
Queued yesterday into the platform-drivers-x86.git’s “for-next” branch are the patches for the Lenovo ThinkPad ACPI driver to begin reporting damaged device detection. This code being in the “for-next” branch makes it material for the next version of the Linux kernel and initially will be able to report to the user on damaged USB-C ports…
Users of specific versions of Microsoft’s Windows 11 can no longer properly shutdown or enter hibernation, and are instead being forced to restart unless the correct Command Prompt command is made, as part of the latest update to Windows 11 Version 23H2. The versions of Windows 11 impacted by the issue are Windows 11 Enterprise and Windows
Claude Cowork, Anthropic’s AI assistant for taking care of simple tasks on your computer, is now available for anyone with a $20 per month Pro subscription to try. Anthropic launched Cowork as an exclusive feature for its Max subscribers, who pay a minimum of $100 per month for more uses of Claude’s expensive reasoning models and early access to experimental features. Now Claude Cowork is available at a cheaper price, though Anthropic notes “Pro users may hit their usage limits earlier” than Max users do.
Like other AI agents, the novelty of Claude Cowork is its ability to work on its own. If you have the macOS Claude app and a Pro subscription, you can prompt Claude Cowork to work on tasks on your local computer, like creating documents based on files you have saved or organizing your folders. The feature is an evolution of Claude Code, Anthropic’s AI coding agent, and can similarly use connectors and the Claude Chrome plugin to work with other apps and the web.
As part of this expanded rollout, Anthropic has included a few fixes inspired by early user feedback. You’ll now be able to rename sessions with Claude Cowork (“Tasks” in the parlance of the Claude app) and the company says the AI assistant will offer better file format previews, more reliable use of connectors to other apps and confirmation messages before it deletes files.
Coding agents top the list of applications of AI that have gained real traction in the last year, so Anthropic applying what it learned with Claude Code to a more general collection of computer tasks makes sense. Claude Cowork is still limited to macOS and Anthropic’s paid subscribers, but assuming the AI agent continues to be popular, it wouldn’t be surprising if the company brought it to other platforms.
This article originally appeared on Engadget at https://www.engadget.com/ai/anthropic-opens-up-its-claude-cowork-feature-to-anyone-with-a-20-subscription-194000021.html?src=rss
Arc Raiders was one of the breakout hits of 2025 and rack up an impressive number of players on Steam. The game has been able to keep its momentum going thanks to a steady stream of content drops, and developers Embark have revealed in an interview with GamesRadar+ that players can look forward to even more this year, including fresh maps.
OpenAI plans to start testing ads inside of ChatGPT “in the coming weeks.” In a blog post published Friday, the company said adult users in the US of its free and Go tiers (more on the latter in a moment) would start seeing sponsored products and services appear below their conversations with its chatbot. “Ads will be clearly labeled and separated from the organic answer,” OpenAI said, adding any sponsored spots would not influence the answers ChatGPT generates. “Answers are optimized based on what’s most helpful to you.”
OpenAI says people won’t see ads appear when they’re talking to ChatGPT about sensitive subjects like their health, mental state of mind or current politics. The company also won’t show ads to teens under the age of 18. As for privacy, OpenAI states it won’t share or sell your data with advertisers. The company will also give users the option to disable ad personalization and clear the data it uses to generate sponsored responses. “We’ll always offer a way to not see ads in ChatGPT, including a paid tier that’s ad-free,” OpenAI adds. Users can dismiss ads, at which point they’ll be asked to explain why they didn’t engage with it.
Users will be able to ask follow-up questions about sponsored content.
OpenAI
“Given what AI can do, we’re excited to develop new experiences over time that people find more helpful and relevant than any other ads. Conversational interfaces create possibilities for people to go beyond static messages and links,” OpenAI said. However, the company was also quick to note its “long-term focus remains on building products that millions of people and businesses find valuable enough to pay for.”
To that point, OpenAI said it would also make its ChatGPT Go subscription available to users in the US. The company first launched the tier in India last August, marketing it as a low-cost alternative to its more expensive Plus and Pro offerings. In the US, Go will cost $8 per month — or $12 less than the monthly price of the Plus plan — and offer 10 times higher rate limits for messages, file uploads and image creation than the free tier. The subscription also extends ChatGPT’s memory and context window, meaning the chatbot will be better at remembering details from past conversations. That said, you’ll see ads at this tier. To go ad-free, you’ll need to subscribe to one of OpenAI’s more expensive plans. For consumers, that means either the Plus or Pro plans.
According to reports, OpenAI had been testing ads inside of ChatGPT since at least the end of last year. As companies continue to pay a high cost for model training and inference, all chatbots are likely to feature ads in some form.
This article originally appeared on Engadget at https://www.engadget.com/ai/openai-is-bringing-ads-to-chatgpt-192831449.html?src=rss
An anonymous reader shares a report: A hit song has been excluded from Sweden’s official chart after it emerged the “artist” behind it was an AI creation. I Know, You’re Not Mine — or Jag Vet, Du Ar Inte Min in Swedish — by a singer called Jacub has been a streaming success in Sweden, topping the Spotify rankings.
However, the Swedish music trade body has excluded the song from the official chart after learning it was AI-generated. “Jacub’s track has been excluded from Sweden’s official chart, Sverigetopplistan, which is compiled by IFPI Sweden. While the song appears on Spotify’s own charts, it does not qualify for inclusion on the official chart under the current rules,” said an IFPI Sweden spokesperson. Ludvig Werber, IFPI Sweden’s chief executive, said: “Our rule is that if it is a song that is mainly AI-generated, it does not have the right to be on the top list.”
Although X removed Grok’s ability to create nonconsensual digitally undressed images on the social platform, the standalone Grok app is another story. It reportedly continues to produce “nudified” deepfakes of real people. And now, Ashley St. Clair, a conservative political strategist and mother of one of Elon Musk’s 14 children, has sued xAI for nonconsensual sexualized images of her that Grok allegedly produced.
In the court filing, St. Clair accused xAI’s Grok chatbot of creating and disseminating deepfakes of her “as a child stripped down to a string bikini, and as an adult in sexually explicit poses, covered in semen, or wearing only bikini floss.” In some cases, the chatbot allegedly produced bikini-clad deepfakes of St. Clair based on a photo of her as a 14-year-old. “People took pictures of me as a child and undressed me. There’s one where they undressed me and bent me over, and in the background is my child’s backpack that he’s wearing right now,” she said.
“I am also seeing images where they add bruises to women, beat them up, tie them up, mutilated,” St. Clair told The Guardian. “These sickos used to have to go to the dark depths of the internet, and now it is on a mainstream social media app.”
St. Clair said that, after she reported the images to X, the social platform replied that the content didn’t violate any policies. In addition, she claims that X left the images posted for up to seven days after she reported them. St. Clair said xAI then retaliated against her by creating more digitally undressed deepfakes of her, therefore “making [St. Clair] the laughingstock of the social media platform.”
She accused the company of then revoking her X Premium subscription, verification checkmark and ability to monetize content on the platform. “xAI further banned [her] from repurchasing Premium,” St. Clair’s court filing states.
On Wednesday, X said it changed its policies so that Grok would no longer generate sexualized images of children or nonconsensual nudity “in those jurisdictions where it’s illegal.” However, the standalone Grok app reportedly continues to undress and sexualize photos when prompted to do so.
Neither Apple nor Google has removed the Grok app despite explicit policy violations.
Anna Moneymaker via Getty Images
Apple and Google have thus far done, well, absolutely nothing. Despite the multi-week outrage over the deepfakes — and an open letter from 28 advocacy groups — neither company has removed the X or Grok apps from their app stores. Both the App Store and Play Store have policies that explicitly prohibit apps that generate such content.
Neither Apple nor Google has responded to multiple requests for comment from Engadget. That includes a follow-up email sent on Friday, regarding the Grok app continuing to “nudify” photos of real women and other people.
“If you are a woman, you can’t post a picture, and you can’t speak, or you risk this abuse,” St. Clair told The Guardian. “It’s dangerous, and I believe this is by design. You are supposed to feed AI humanity and thoughts, and when you are doing things that particularly impact women, and they don’t want to participate in it because they are being targeted, it means the AI is inherently going to be biased.”
Speaking about Musk and his team, she added that “these people believe they are above the law, because they are. They don’t think they are going to get in trouble, they think they have no consequences.”
This article originally appeared on Engadget at https://www.engadget.com/ai/the-mother-of-one-of-elon-musks-children-is-suing-xai-over-nonconsensual-deepfake-images-191451979.html?src=rss
Verizon has received all approvals it needs for a $9.6 billion acquisition of Frontier Communications, an Internet service provider with about 3.3 million broadband customers in 25 states. Verizon said it expects to complete the merger on January 20.
The last approval came from the California Public Utilities Commission (CPUC), which allowed the deal in a 5–0 vote yesterday. There were months of negotiations that resulted in requirements to deploy more fiber and wireless infrastructure, offer $20-per-month Internet service to people with low incomes for the next decade, and other commitments, including some designed to replace the DEI (diversity, equity, and inclusion) policies that Verizon had to end because of demands by the Trump administration.
“The approval follows extensive public participation, testimony from multiple parties, and negotiated settlement agreements with consumer advocates and labor organizations,” the CPUC said yesterday.
We may earn a commission from links on this page. Deal pricing and availability subject to change after time of publication.
If you’re dreaming of ditching a multi-monitor gaming setup without sacrificing too much screen space, the Samsung 49-inch ViewFinity S95UCmonitordelivers that idea in a single curved panel. A versatile option for both multi-tasking at work and casual gaming, it’s currently 33% off at $799 (originally $1,199).
This entry-level gaming monitor is ultra-wide (32:9) with a Dual QHD display (5,120 x 1,440) with HDR400 support, a height-adjustable stand, and built-in speakers, which maximize your desk space even more. While its giant display with tons of usable space is the main draw, its 120Hz refresh rate offers smooth visuals and fast reactions, but may not suffice for pro-gamers who need an ultra-fast 240Hz-plus performance for competitive play.
VESA DisplayHDR 400 gives you solid color accuracy and brightness, though it can’t match the contrast of an OLED panel, and users may experience a slight gamma shift at the edges of this panel due to its width. USB-C with up to 90 W charging and a KVM Switch, which allows users to connect and control two sources to the display at the same time from a single keyboard and mouse, makes it an effective dock-style hub for laptops and desktops.
If you’re looking more screen real estate in a single monitor, smoother visuals than a basic office monitor, and an option that you can toggle between both during work and after hours for your gaming needs, the Samsung 49-inch ViewFinity S95UC monitor may not be the fastest or HDR-impressive compared to higher-end OLED models. But at around $800, this ultra-wide display is a capable, all-in-one alternative to a dual-monitor setup that also helps you reclaim more desk space.
Zwift Racing League teams headed to Yorkshire this week for a points race held on the 2019 UCI Worlds Harrogate Circuit. Over the years, this route’s sawtooth profile has struck fear into the heart of many a Zwifter, including myself. And our race was three laps long, meaning it would take us over an hour to finish.
My team (Coalition Delusion, racing in the Open Development B1 Lime Division) knew this wouldn’t be easy, and it wouldn’t be short. But you’re not a true cyclist unless you regularly run pell-mell toward the prospect of personal suffering. Let’s race!
Lap 1
73 riders left the Yorkshire start pens, quickly turning right to begin climbing Otley Road. (You’ll want to be nicely warmed up before racing in Yorkshire, as the work begins right away!) Otley Road is 1.6km long at 3.4%, a draggy climb that always seems to take longer than you’d think.
But this was a points race, and Otley Road isn’t a timed segment. With plenty of miles ahead, riders were clearly wanting to keep their powder dry. I sat in the wheels, doing the minimal work to stay in the group. I finished the climb in 3:22, averaging 3.96 W/kg.
What follows on the route is a bit of down, a bit of up, some false flat, then the lovely Pot Bank descent followed by another short climb and a descent to Oak Beck, where the Yorkshire KOM begins. Interesting things can happen on these in-between bits, but in a points race, that action is muted.
We hit the bottom of the KOM as a pack of 64, and I bumped up my power and used my feather powerup to stay near (but not on) the front if possible. New teammate Enrico was riding off the front (impressive!), so I eased even more, not wanting to give anyone a draft to chase him back (insert timely Team Jayco-AlUla reference ).
At the flatter section mid-climb I was positioned well, but when the second half began, lots of feathers popped, and I found myself sliding backward. By the time I rode beneath the KOM arch I had averaged 4.64 W/kg for 2:32 (tying my PR time), the pack was strung out, and I was in… well, I really don’t know what place I was in.
The rider list on my screen (see above) wasn’t showing my current place in the race, and I must say, I didn’t realize how much I looked at that HUD element until it was missing!
I had Sauce for Zwift up, which showed me in a group of 29… but that was a very strung-out group that was quickly breaking up.
(After reporting the bug via an internal Zwift channel, it appears that this happened to anyone running the latest Zwift version, v105, that released the morning of the race. The new version interacts with ZRL’s funky event config in an unexpected way.)
I figured the groups would come back together once we were through the sprint segment, and that is indeed what happened. By the time we finished the first lap, I was back in the front group of 28. With a 31-second gap to the closest chasing group of 5 riders, it seemed pretty clear that nobody behind would be bridging up.
Lap 2
The next Otley Road ascent was a bit more chill, taking 3:35 and averaging 3.57 W/kg. Nothing remarkable here, except for one Team Ukraine Mariupol guy, Klish, who attacked off the front. He would continue to do this on and off for the rest of the race, going off the front, getting caught, then doing it again. I’m not sure what the strategy was, but without a lot of TUM guys in the front group, I don’t think anyone was worried enough to chase him down in earnest.
As we neared the Pot Bank descent this time, I decided to put in a few hard pedal revs to go off the front and attempt a supertuck on the crazy steep descent. And it worked!
That gave me almost 20 seconds of no pedaling, which was quite nice.
Soon enough, we were at the start of our second Yorkshire KOM. This time, I had determined to save my feather powerup for the second half of the climb, where the road is just as steep as the start of the climb, but you’re more fatigued! Clearly others had learned the same lesson, as I saw zero feathers at the start, but a pile on the second bit:
This time up the KOM was definitely easier than the first. We were 8 seconds slower, and it took me 0.16 W/kg less, and I finished the climb in a much better position than the first time around, as well.
This is where my glycogen-depleted brain began to dream that I might be able to hang with the front pack to the finish. But those hopes were soon to come up against the hard wall of reality…
I would love to say that I was able to contest the sprint segment each lap, but since it was at the top of a climb, I was always gassed by the time I got there, and was only able to put in half-hearted attempts. I just hoped that my being in the whittled-down front pack would earn sufficient points.
We finished lap 2 with a front group of 18 riders.
Lap 3
As we climbed Otley Road for the final time, I was surveying the front group. How many riders were here from each team? I was happy to see that most teams only had 1-2 riders in the front, while my team and one other (SZR) had 3. That was a good sign.
We finished Otley Road in 4:01 this time, much slower than the previous laps. Fatigue was setting in, and everyone knew the final KOM would be a hard one!
I grabbed another cheeky supertuck on the Pot Bank descent, and soon enough, we found ourselves crossing Oak Beck and beginning the final KOM. It felt like the group was pushing hard up the first half, and I found myself sliding backward in the group. By the time we started the second half I was blowing up. It wasn’t that the pace was higher than previous laps. It was that my legs didn’t have any punch left. Even my feather powerup couldn’t save me.
I was dropped. 17 riders were up ahead, and I was all by myself, with the closest riders over a minute behind.
Doing whatever I could to keep pushing over the top and down the other side, I used my aero powerup on the descent in hopes that the pack would ease and I could catch back on. One rider had fallen off the group, and I caught and passed him on the climb to the sprint. (He would serve as a sort of “reverse carrot”, forcing me to keep my effort high to stay away to the finish.)
Alas, the pack never eased in those final kilometers. Teammate Andrew finished in 2nd, the best finish for our team (chapeau!), new teammate Enrico finished 5th, and I finished in 17th place, just over a minute behind the leaders.
On the plus side, I didn’t even need to sprint, because nobody was nearby.
It took several hours for the results to be finalized (not sure why it takes longer sometimes), but eventually we saw we’d finished 4th overall:
Our rivals from last season, Team SEA, took first. They’re sitting in 1st overall after two races, and we’re in 2nd.
But here’s the thing: we only had 5 riders. So we’re pretty happy with 4th this week, knowing that having one more rider finishing mid-pack might have won it for us overall. That’s not bad at all, on such a hilly route.
Personally, I’m both happy and disappointed in my performance. On one hand, I didn’t think I’d survive in the front as long as I did. But on the other hand, getting dropped so close to the finish really stings! I can’t help but wonder if using my feather earlier, or gutting it out just a bit more, or being able to see my rider placing might have helped me stay with the front to the finish.
The truth is, I probably just wasn’t strong enough on the day. But I reported the bug to Zwift via a Slack channel anyway, asking, “Can I use this to excuse my getting dropped like a hot rock on the last KOM? Please?” Happily, Zwift’s VP of Product Mark Cote said I could, and admitted that Zwift was clearly at fault for my loss:
So that helps lessen the sting just a bit.
We finished with our customary Discord team portrait, featuring “shadow-Captain Neil” (at bottom) who honorably did this event twice in one day:
Last week at CES, Lego introduced its new Smart Play system, with a tech-packed Smart Brick that can recognize and interact with sets and minifigures. It was unexpected and delightful to see Lego come up with a way to modernize its bricks without the need for apps, screens or AI.
So I was a little surprised this week when the Lego Education group announced its latest initiative is the Computer Science and AI Learning Solution. After all, generative AI feels like the antithesis of Lego’s creative values. But Andrew Silwinski, Lego Education’s head of product experience, was quick to defend Lego’s approach, noting that being fluent in the tools behind AI is not about generating sloppy images or music and more about expanding what it means by teaching computer science.
“I think most people should probably know that we started working on this before ChatGPT [got big],” Silwinski told Engadget earlier this week. “Some of the ideas that underline AI are really powerful foundational ideas, regardless of the current frontier model that’s out this week. Helping children understand probability and statistics, data quality, algorithmic bias, sensors, machine perception. These are really foundational core ideas that go back to the 1970s.”
To that end, Lego Education designed courses for grades K-2, 3-5 and 6-8 that incorporate Lego bricks, additional hardware and lessons tailored to introducing the fundamentals of AI as an extension of existing computer science education. The kits are designed for four students to work together, with teacher oversight. Much of this all comes from learnings Lego found in a study it commissioned showing that teachers often find they don’t have the right resources to teach these subjects. The study showed that half of teachers globally say “current resources leave students bored” while nearly half say “computer science isn’t relatable and doesn’t connect to students’ interests or day to day.” Given kids’ familiarity with Lego and the multiple decades of experience Lego Education has in putting courses like this together, it seems like a logical step to push in this direction.
In Lego’s materials about the new courses, AI is far from the only subject covered. Coding, looping code, triggering events and sequences, if/then conditionals and more are all on display through the combination of Lego-built models and other hardware to motorize it. It feels more like a computer science course that also introduces concepts of AI rather than something with an end goal of having kids build a chatbot.
In fact, Lego set up a number of “red lines” in terms of how it would introduce AI. “No data can ever go across the internet to us or any other third party,” Silwinski said. “And that’s a really hard bar if you know anything about AI.” So instead of going to the cloud, everything had to be able to do local inference on, as Silwinski said, “the 10-year-old Chromebooks you’ll see in classrooms.” He added that “kids can train their own machine learning models, and all of that is happening locally in the classroom, and none of that data ever leaves the student’s device.”
Lego also says that its lessons never anthropomorphize AI, one of the things that is so common in consumer-facing AI tools like ChatGPT, Gemini and many more. “One of the things we’re seeing a lot of with generative AI tools is children have a tendency to see them as somehow human or almost magical. A lot of it’s because of the conversational interface, it abstracts all the mechanics away from the child.”
Lego also recognized that it had to build a course that’ll work regardless of a teacher’s fluency in such subjects. So a big part of developing the course was making sure that teachers had the tools they needed to be on top of whatever lessons they’re working on. “When we design and we test the products, we’re not the ones testing in the classroom,” Silwinski said. “We give it to a teacher and we provide all of the lesson materials, all of the training, all of the notes, all the presentation materials, everything that they need to be able to teach the lesson.” Lego also took into account the fact that some schools might introduce its students to these things starting in Kindergarten, whereas others might skip to the grade 3-5 or 6-8 sets. To alleviate any bumps in the courses for students or teachers, Lego Education works with school districts and individual schools to make sure there’s an on-ramp for those starting from different places in their fluency.
While the idea of “teaching AI” seemed out of character for Lego initially, the approach it’s taking here actually reminds me a bit of Smart Play. With Smart Play, the technology is essentially invisible — kids can just open up a set, start building, and get all the benefits of the new system without having to hook up to an app or a screen. In the same vein, Silwinski said that a lot of the work you can do with the Computer Science and AI kit doesn’t need a screen, particularly the lessons designed for younger kids. And the sets themselves have a mode that acts similar to a mesh, where you connect numerous motors and sensors together to build “incredibly complex interactions and behaviors” without even needing a computer.
For educators interested in checking out this latest course, Lego has single kits up for pre-order starting at $339.95; they’ll start shipping in April. That’s the pricing for the K-2 sets, the 3-5 and 6-8 sets are $429.95 and $529.95, respectively. A single kit covers four students. Lego is also selling bundles with six kits, and school districts can also request a quote for bigger orders.
This article originally appeared on Engadget at https://www.engadget.com/ai/legos-latest-educational-kit-seeks-to-teach-ai-as-part-of-computer-science-not-to-build-a-chatbot-184636741.html?src=rss