Tech

backblaze-responds-to-claims-of-“sham-accounting,”-customer-backups-at-risk

Backblaze responds to claims of “sham accounting,” customer backups at risk

Backblaze went public in November 2021 and raised $100 million. Morpheus noted that since then, “Backblaze has reported losses every quarter, its outstanding share count has grown by 80 percent, and its share price has declined by 71 percent.”

Following Morpheus’ report, Investing reported on Thursday that Backblaze shares fell 8.3 percent.

Beyond the financial implications for stockholders, Morpheus’ report has sparked some concern for the primarily small businesses and individuals relying on Backblaze for data backup. Today, for example, How-To Geek reported that “Backblaze backups might be in trouble,” in reference to Morpheus’ report. The publication said that if Morpheus’ reporting was accurate, Backblaze doesn’t appear to be heading toward profitability. In its Q4 2024 earnings report [PDF], Backblaze reported a net loss of $48.5 million. In 2023, it reported a net loss of $59.7 million.

“If Backblaze suddenly shuts down, customers might lose access to existing backups,” How-To Geek said.

Backblaze responds

Ars Technica reached out to Backblaze about its response to concerns about the company’s financials resulting in lost backups. Patrick Thomas, Backblaze’s VP of marketing, called Morpheus’ claims “baseless.” He added:

The report is inaccurate and misleading, based largely on litigation of the same nature, and a clear attempt by short sellers to manipulate our stock price for financial gain.

Thomas also claimed that “independent, third-party reviews” have already found that there have been “no wrongdoing or issues” with Backblaze’s public financial results.

“Our storage cloud continues to deliver reliable, high-performance services that Backblaze customers rely on, and we remain fully focused on driving innovation and creating long-term value for our customers, employees, and investors,” Thomas said.

Backblaze will announce its Q1 2025 results on May 7. Regardless of what lies ahead for the company’s finances and litigation, commitment to the 3-2-1 backup rule remains prudent.

Backblaze responds to claims of “sham accounting,” customer backups at risk Read More »

ios-and-android-juice-jacking-defenses-have-been-trivial-to-bypass-for-years

iOS and Android juice jacking defenses have been trivial to bypass for years


SON OF JUICE JACKING ARISES

New ChoiceJacking attack allows malicious chargers to steal data from phones.

Credit: Aurich Lawson | Getty Images

Credit: Aurich Lawson | Getty Images

About a decade ago, Apple and Google started updating iOS and Android, respectively, to make them less susceptible to “juice jacking,” a form of attack that could surreptitiously steal data or execute malicious code when users plug their phones into special-purpose charging hardware. Now, researchers are revealing that, for years, the mitigations have suffered from a fundamental defect that has made them trivial to bypass.

“Juice jacking” was coined in a 2011 article on KrebsOnSecurity detailing an attack demonstrated at a Defcon security conference at the time. Juice jacking works by equipping a charger with hidden hardware that can access files and other internal resources of phones, in much the same way that a computer can when a user connects it to the phone.

An attacker would then make the chargers available in airports, shopping malls, or other public venues for use by people looking to recharge depleted batteries. While the charger was ostensibly only providing electricity to the phone, it was also secretly downloading files or running malicious code on the device behind the scenes. Starting in 2012, both Apple and Google tried to mitigate the threat by requiring users to click a confirmation button on their phones before a computer—or a computer masquerading as a charger—could access files or execute code on the phone.

The logic behind the mitigation was rooted in a key portion of the USB protocol that, in the parlance of the specification, dictates that a USB port can facilitate a “host” device or a “peripheral” device at any given time, but not both. In the context of phones, this meant they could either:

  • Host the device on the other end of the USB cord—for instance, if a user connects a thumb drive or keyboard. In this scenario, the phone is the host that has access to the internals of the drive, keyboard or other peripheral device.
  • Act as a peripheral device that’s hosted by a computer or malicious charger, which under the USB paradigm is a host that has system access to the phone.

An alarming state of USB security

Researchers at the Graz University of Technology in Austria recently made a discovery that completely undermines the premise behind the countermeasure: They’re rooted under the assumption that USB hosts can’t inject input that autonomously approves the confirmation prompt. Given the restriction against a USB device simultaneously acting as a host and peripheral, the premise seemed sound. The trust models built into both iOS and Android, however, present loopholes that can be exploited to defeat the protections. The researchers went on to devise ChoiceJacking, the first known attack to defeat juice-jacking mitigations.

“We observe that these mitigations assume that an attacker cannot inject input events while establishing a data connection,” the researchers wrote in a paper scheduled to be presented in August at the Usenix Security Symposium in Seattle. “However, we show that this assumption does not hold in practice.”

The researchers continued:

We present a platform-agnostic attack principle and three concrete attack techniques for Android and iOS that allow a malicious charger to autonomously spoof user input to enable its own data connection. Our evaluation using a custom cheap malicious charger design reveals an alarming state of USB security on mobile platforms. Despite vendor customizations in USB stacks, ChoiceJacking attacks gain access to sensitive user files (pictures, documents, app data) on all tested devices from 8 vendors including the top 6 by market share.

In response to the findings, Apple updated the confirmation dialogs in last month’s release of iOS/iPadOS 18.4 to require a user authentication in the form of a PIN or password. While the researchers were investigating their ChoiceJacking attacks last year, Google independently updated its confirmation with the release of version 15 in November. The researchers say the new mitigation works as expected on fully updated Apple and Android devices. Given the fragmentation of the Android ecosystem, however, many Android devices remain vulnerable.

All three of the ChoiceJacking techniques defeat Android juice-jacking mitigations. One of them also works against those defenses in Apple devices. In all three, the charger acts as a USB host to trigger the confirmation prompt on the targeted phone.

The attacks then exploit various weaknesses in the OS that allow the charger to autonomously inject “input events” that can enter text or click buttons presented in screen prompts as if the user had done so directly into the phone. In all three, the charger eventually gains two conceptual channels to the phone: (1) an input one allowing it to spoof user consent and (2) a file access connection that can steal files.

An illustration of ChoiceJacking attacks. (1) The victim device is attached to the malicious charger. (2) The charger establishes an extra input channel. (3) The charger initiates a data connection. User consent is needed to confirm it. (4) The charger uses the input channel to spoof user consent. Credit: Draschbacher et al.

It’s a keyboard, it’s a host, it’s both

In the ChoiceJacking variant that defeats both Apple- and Google-devised juice-jacking mitigations, the charger starts as a USB keyboard or a similar peripheral device. It sends keyboard input over USB that invokes simple key presses, such as arrow up or down, but also more complex key combinations that trigger settings or open a status bar.

The input establishes a Bluetooth connection to a second miniaturized keyboard hidden inside the malicious charger. The charger then uses the USB Power Delivery, a standard available in USB-C connectors that allows devices to either provide or receive power to or from the other device, depending on messages they exchange, a process known as the USB PD Data Role Swap.

A simulated ChoiceJacking charger. Bidirectional USB lines allow for data role swaps. Credit: Draschbacher et al.

With the charger now acting as a host, it triggers the file access consent dialog. At the same time, the charger still maintains its role as a peripheral device that acts as a Bluetooth keyboard that approves the file access consent dialog.

The full steps for the attack, provided in the Usenix paper, are:

1. The victim device is connected to the malicious charger. The device has its screen unlocked.

2. At a suitable moment, the charger performs a USB PD Data Role (DR) Swap. The mobile device now acts as a USB host, the charger acts as a USB input device.

3. The charger generates input to ensure that BT is enabled.

4. The charger navigates to the BT pairing screen in the system settings to make the mobile device discoverable.

5. The charger starts advertising as a BT input device.

6. By constantly scanning for newly discoverable Bluetooth devices, the charger identifies the BT device address of the mobile device and initiates pairing.

7. Through the USB input device, the charger accepts the Yes/No pairing dialog appearing on the mobile device. The Bluetooth input device is now connected.

8. The charger sends another USB PD DR Swap. It is now the USB host, and the mobile device is the USB device.

9. As the USB host, the charger initiates a data connection.

10. Through the Bluetooth input device, the charger confirms its own data connection on the mobile device.

This technique works against all but one of the 11 phone models tested, with the holdout being an Android device running the Vivo Funtouch OS, which doesn’t fully support the USB PD protocol. The attacks against the 10 remaining models take about 25 to 30 seconds to establish the Bluetooth pairing, depending on the phone model being hacked. The attacker then has read and write access to files stored on the device for as long as it remains connected to the charger.

Two more ways to hack Android

The two other members of the ChoiceJacking family work only against the juice-jacking mitigations that Google put into Android. In the first, the malicious charger invokes the Android Open Access Protocol, which allows a USB host to act as an input device when the host sends a special message that puts it into accessory mode.

The protocol specifically dictates that while in accessory mode, a USB host can no longer respond to other USB interfaces, such as the Picture Transfer Protocol for transferring photos and videos and the Media Transfer Protocol that enables transferring files in other formats. Despite the restriction, all of the Android devices tested violated the specification by accepting AOAP messages sent, even when the USB host hadn’t been put into accessory mode. The charger can exploit this implementation flaw to autonomously complete the required user confirmations.

The remaining ChoiceJacking technique exploits a race condition in the Android input dispatcher by flooding it with a specially crafted sequence of input events. The dispatcher puts each event into a queue and processes them one by one. The dispatcher waits for all previous input events to be fully processed before acting on a new one.

“This means that a single process that performs overly complex logic in its key event handler will delay event dispatching for all other processes or global event handlers,” the researchers explained.

They went on to note, “A malicious charger can exploit this by starting as a USB peripheral and flooding the event queue with a specially crafted sequence of key events. It then switches its USB interface to act as a USB host while the victim device is still busy dispatching the attacker’s events. These events therefore accept user prompts for confirming the data connection to the malicious charger.”

The Usenix paper provides the following matrix showing which devices tested in the research are vulnerable to which attacks.

The susceptibility of tested devices to all three ChoiceJacking attack techniques. Credit: Draschbacher et al.

User convenience over security

In an email, the researchers said that the fixes provided by Apple and Google successfully blunt ChoiceJacking attacks in iPhones, iPads, and Pixel devices. Many Android devices made by other manufacturers, however, remain vulnerable because they have yet to update their devices to Android 15. Other Android devices—most notably those from Samsung running the One UI 7 software interface—don’t implement the new authentication requirement, even when running on Android 15. The omission leaves these models vulnerable to ChoiceJacking. In an email, principal paper author Florian Draschbacher wrote:

The attack can therefore still be exploited on many devices, even though we informed the manufacturers about a year ago and they acknowledged the problem. The reason for this slow reaction is probably that ChoiceJacking does not simply exploit a programming error. Rather, the problem is more deeply rooted in the USB trust model of mobile operating systems. Changes here have a negative impact on the user experience, which is why manufacturers are hesitant. [It] means for enabling USB-based file access, the user doesn’t need to simply tap YES on a dialog but additionally needs to present their unlock PIN/fingerprint/face. This inevitably slows down the process.

The biggest threat posed by ChoiceJacking is to Android devices that have been configured to enable USB debugging. Developers often turn on this option so they can troubleshoot problems with their apps, but many non-developers enable it so they can install apps from their computer, root their devices so they can install a different OS, transfer data between devices, and recover bricked phones. Turning it on requires a user to flip a switch in Settings > System > Developer options.

If a phone has USB Debugging turned on, ChoiceJacking can gain shell access through the Android Debug Bridge. From there, an attacker can install apps, access the file system, and execute malicious binary files. The level of access through the Android Debug Mode is much higher than that through Picture Transfer Protocol and Media Transfer Protocol, which only allow read and write access to system files.

The vulnerabilities are tracked as:

    • CVE-2025-24193 (Apple)
    • CVE-2024-43085 (Google)
    • CVE-2024-20900 (Samsung)
    • CVE-2024-54096 (Huawei)

A Google spokesperson confirmed that the weaknesses were patched in Android 15 but didn’t speak to the base of Android devices from other manufacturers, who either don’t support the new OS or the new authentication requirement it makes possible. Apple declined to comment for this post.

Word that juice-jacking-style attacks are once again possible on some Android devices and out-of-date iPhones is likely to breathe new life into the constant warnings from federal authorities, tech pundits, news outlets, and local and state government agencies that phone users should steer clear of public charging stations.

As I reported in 2023, these warnings are mostly scaremongering, and the advent of ChoiceJacking does little to change that, given that there are no documented cases of such attacks in the wild. That said, people using Android devices that don’t support Google’s new authentication requirement may want to refrain from public charging.

Photo of Dan Goodin

Dan Goodin is Senior Security Editor at Ars Technica, where he oversees coverage of malware, computer espionage, botnets, hardware hacking, encryption, and passwords. In his spare time, he enjoys gardening, cooking, and following the independent music scene. Dan is based in San Francisco. Follow him at here on Mastodon and here on Bluesky. Contact him on Signal at DanArs.82.

iOS and Android juice jacking defenses have been trivial to bypass for years Read More »

“you-wouldn’t-steal-a-car”-anti-piracy-campaign-may-have-used-pirated-fonts

“You wouldn’t steal a car” anti-piracy campaign may have used pirated fonts

Aquilina, who was speaking generally and not on the specifics of the anti-piracy campaign and its font use, said that using a font from a free source, with an “effectively implied license to use it,” could be “a good defense,” though “not a complete defense.” Typically, a rightsholder would go after websites distributing copies of their font, not after users of the end product.

Fonts used commercially that happen to be exact copies of existing and copyrighted fonts are “fairly common,” Aquilina said, “simply because of the popularity of certain fonts and a desire to use them, to create a certain aesthetic.” But, he said, there is “a very small percentage that could be, or are, litigated.” Even with software licenses at issue, a type foundry faces an uphill battle, as witnessed in the battle over Shake Shack’s typography (paywalled).

Still missing: the source of XBand Rough

A few glyphs from FF Confidential, the font that was not used on some anti-piracy materials, even if it sure looked like that.

A few glyphs from FF Confidential, the font that was not used on some anti-piracy materials, even if it sure looked like that. Credit: MyFonts/MonotType

So where did Xband Rough come from?

The styling of the font name, “XBAND Rough” with the first noun in all-caps, calls to mind the early online gaming network XBAND, launched in 1994 and discontinued in 1997. In some XBand packages, a similar “rough” style can be seen on the lettering. The PDF sleuth, Rib, noted that XBAND Rough “came out four years after the original” (about 1996) and was “near-identical, except for the price.”

Another Bluesky user suggests “a plausible explanation” for the font, suggesting that Xband may have licensed FF Confidential and then given it the internal name “Xband Rough.” A copy of the font with that name could have been extracted from some Xband material and then “started floating around the Internet uncredited.” In the end, though, the real answer is unclear.

We contacted the Motion Picture Association (now just the MPA, sans “of America”), but they declined to comment.

The original “You Wouldn’t Steal a Car” campaign was simple to the point of being simplistic. IP law isn’t really like “stealing a car” in many cases—as has made clearly once again by the recent Xband Rough investigation.

“You wouldn’t steal a car” anti-piracy campaign may have used pirated fonts Read More »

perplexity-will-come-to-moto-phones-after-exec-testified-google-limited-access

Perplexity will come to Moto phones after exec testified Google limited access

Shevelenko was also asked about Chrome, which the DOJ would like to force Google to sell. Like an OpenAI executive said on Monday, Shevelenko confirmed Perplexity would be interested in buying the browser from Google.

Motorola has all the AI

There were some vague allusions during the trial that Perplexity would come to Motorola phones this year, but we didn’t know just how soon that was. With the announcement of its 2025 Razr devices, Moto has confirmed a much more expansive set of AI features. Parts of the Motorola AI experience are powered by Gemini, Copilot, Meta, and yes, Perplexity.

While Gemini gets top billing as the default assistant app, other firms have wormed their way into different parts of the software. Perplexity’s app will be preloaded, and anyone who buys the new Razrs. Owners will also get three free months of Perplexity Pro. This is the first time Perplexity has had a smartphone distribution deal, but it won’t be shown prominently on the phone. When you start a Motorola device, it will still look like a Google playground.

While it’s not the default assistant, Perplexity is integrated into the Moto AI platform. The new Razrs will proactively suggest you perform an AI search when accessing certain features like the calendar or browsing the web under the banner “Explore with Perplexity.” The Perplexity app has also been optimized to work with the external screen on Motorola’s foldables.

Moto AI also has elements powered by other AI systems. For example, Microsoft Copilot will appear in Moto AI with an “Ask Copilot” option. And Meta’s Llama model powers a Moto AI feature called Catch Me Up, which summarizes notifications from select apps.

It’s unclear why Motorola leaned on four different AI providers for a single phone. It probably helps that all these companies are desperate to entice users to bulk up their market share. Perplexity confirmed that no money changed hands in this deal—it’s on Moto phones to acquire more users. That might be tough with Gemini getting priority placement, though.

Perplexity will come to Moto phones after exec testified Google limited access Read More »

roku-tech,-patents-prove-its-potential-for-delivering-“interruptive”-ads

Roku tech, patents prove its potential for delivering “interruptive” ads

Roku, owner of one of the most popular connected TV operating systems in the country, walks a fine line when it comes to advertising. Roku’s OS lives on low-priced smart TVs, streaming sticks, and projectors. To make up the losses from cheaply priced hardware, Roku is dependent on selling advertisements throughout its OS, including screensavers and its home screen.

That business model has pushed Roku to experiment with new ways of showing ads that test users’ tolerance. The company claims that it doesn’t want ads on its platform to be considered intrusive, but there are reasons to be skeptical about Roku’s pledge.

Non-“interruptive” ads

In an interview with The Verge this week, Jordan Rost, Roku’s head of ad marketing, emphasized that Roku tries to only deliver ads that don’t interrupt viewers.

“Advertisers want to be part of a good experience. They don’t want to be interruptive,” he told The Verge.

Rost noted that Roku is always testing new ad formats. Those tests include doing “all of our own A/B testing on the platform” and listening to customer feedback, he added.

“We’re constantly tweaking and trying to figure out what’s going to be helpful for the user experience,” Rost said.

For many streamers, however, ads and a better user experience are contradictory. In fact, for many, the simplest way to improve streaming is fewer ads and a more streamlined access to content. That’s why Apple TV boxes, which doesn’t have integrated ads and is good at combining content from multiple streaming subscriptions, is popular among Ars Technica staff and readers. An aversion to ads is also why millions pay extra for ad-free streaming subscriptions.

Roku tech, patents prove its potential for delivering “interruptive” ads Read More »

review:-ryzen-ai-cpu-makes-this-the-fastest-the-framework-laptop-13-has-ever-been

Review: Ryzen AI CPU makes this the fastest the Framework Laptop 13 has ever been


With great power comes great responsibility and subpar battery life.

The latest Framework Laptop 13, which asks you to take the good with the bad. Credit: Andrew Cunningham

The latest Framework Laptop 13, which asks you to take the good with the bad. Credit: Andrew Cunningham

At this point, the Framework Laptop 13 is a familiar face, an old friend. We have reviewed this laptop five other times, and in that time, the idea of a repairable and upgradeable laptop has gone from a “sounds great if they can pull it off” idea to one that’s become pretty reliable and predictable. And nearly four years out from the original version—which shipped with an 11th-generation Intel Core processor—we’re at the point where an upgrade will get you significant boosts to CPU and GPU performance, plus some other things.

We’re looking at the Ryzen AI 300 version of the Framework Laptop today, currently available for preorder and shipping in Q2 for people who buy one now. The laptop starts at $1,099 for a pre-built version and $899 for a RAM-less, SSD-less, Windows-less DIY version, and we’ve tested the Ryzen AI 9 HX 370 version that starts at $1,659 before you add RAM, an SSD, or an OS.

This board is a direct upgrade to Framework’s Ryzen 7040-series board from mid-2023, with most of the same performance benefits we saw last year when we first took a look at the Ryzen AI 300 series. It’s also, if this matters to you, the first Framework Laptop to meet Microsoft’s requirements for its Copilot+ PC initiative, giving users access to some extra locally processed AI features (including but not limited to Recall) with the promise of more to come.

For this upgrade, Ryzen AI giveth, and Ryzen AI taketh away. This is the fastest the Framework Laptop 13 has ever been (at least, if you spring for the Ryzen AI 9 HX 370 chip that our review unit shipped with). If you’re looking to do some light gaming (or non-Nvidia GPU-accelerated computing), the Radeon 890M GPU is about as good as it gets. But you’ll pay for it in battery life—never a particularly strong point for Framework, and less so here than in most of the Intel versions.

What’s new, Framework?

This Framework update brings the return of colorful translucent accessories, parts you can also add to an older Framework Laptop if you want. Credit: Andrew Cunningham

We’re going to focus on what makes this particular Framework Laptop 13 different from the past iterations. We talk more about the build process and the internals in our review of the 12th-generation Intel Core version, and we ran lots of battery tests with the new screen in our review of the Intel Core Ultra version. We also have coverage of the original Ryzen version of the laptop, with the Ryzen 7 7840U and Radeon 780M GPU installed.

Per usual, every internal refresh of the Framework Laptop 13 comes with another slate of external parts. Functionally, there’s not a ton of exciting stuff this time around—certainly nothing as interesting as the higher-resolution 120 Hz screen option we got with last year’s Intel Meteor Lake update—but there’s a handful of things worth paying attention to.

Functionally, Framework has slightly improved the keyboard, with “a new key structure” on the spacebar and shift keys that “reduce buzzing when your speakers are cranked up.” I can’t really discern a difference in the feel of the keyboard, so this isn’t a part I’d run out to add to my own Framework Laptop, but it’s a fringe benefit if you’re buying an all-new laptop or replacing your keyboard for some other reason.

Keyboard legends have also been tweaked; pre-built Windows versions get Microsoft’s dedicated (and, within limits, customizable) Copilot key, while DIY editions come with a Framework logo on the Windows/Super key (instead of the word “super”) and no Copilot key.

Cosmetically, Framework is keeping the dream of the late ’90s alive with translucent plastic parts, namely the bezel around the display and the USB-C Expansion Modules. I’ll never say no to additional customization options, though I still think that “silver body/lid with colorful bezel/ports” gives the laptop a rougher, unfinished-looking vibe.

Like the other Ryzen Framework Laptops (both 13 and 16), not all of the Ryzen AI board’s four USB-C ports support all the same capabilities, so you’ll want to arrange your ports carefully.

Framework’s recommendations for how to configure the Ryzen AI laptop’s expansion modules. Credit: Framework

Framework publishes a graphic to show you which ports do what; if you’re looking at the laptop from the front, ports 1 and 3 are on the back, and ports 2 and 4 are toward the front. Generally, ports 1 and 3 are the “better” ones, supporting full USB4 speeds instead of USB 3.2 and DisplayPort 2.0 instead of 1.4. But USB-A modules should go in ports 2 or 4 because they’ll consume extra power in bays 1 and 3. All four do support display output, though, which isn’t the case for the Ryzen 7040 Framework board, and all four continue to support USB-C charging.

The situation has improved from the 7040 version of the Framework board, where not all of the ports could do any kind of display output. But it still somewhat complicates the laptop’s customizability story relative to the Intel versions, where any expansion card can go into any port.

I will also say that this iteration of the Framework laptop hasn’t been perfectly stable for me. The problems are intermittent but persistent, despite using the latest BIOS version (3.03 as of this writing) and driver package available from Framework. I had a couple of total-system freezes/crashes, occasional problems waking from sleep, and sporadic rendering glitches in Microsoft Edge. These weren’t problems I’ve had with the other Ryzen AI laptops I’ve used so far or with the Ryzen 7040 version of the Framework 13. They also persisted across two separate clean installs of Windows.

It’s possible/probable that some combination of firmware and driver updates can iron out these problems, and they generally didn’t prevent me from using the laptop the way I wanted to use it, but I thought it was worth mentioning since my experience with new Framework boards has usually been a bit better than this.

Internals and performance

“Ryzen AI” is AMD’s most recent branding update for its high-end laptop chips, but you don’t actually need to care about AI to appreciate the solid CPU and GPU speed upgrades compared to the last-generation Ryzen Framework or older Intel versions of the laptop.

Our Framework Laptop board uses the fastest processor offering: a Ryzen AI 9 HX 370 with four of AMD’s Zen 5 CPU cores, eight of the smaller, more power-efficient Zen 5c cores, and a Radeon 890M integrated GPU with 16 of AMD’s RDNA 3.5 graphics cores.

There are places where the Intel Arc graphics in the Core Ultra 7/Meteor Lake version of the Framework Laptop are still faster than what AMD can offer, though your experience may vary depending on the games or apps you’re trying to use. Generally, our benchmarks show the Arc GPU ahead by a small amount, but it’s not faster across the board.

Relative to other Ryzen AI systems, the Framework Laptop’s graphics performance also suffers somewhat because socketed DDR5 DIMMs don’t run as fast as RAM that’s been soldered to the motherboard. This is one of the trade-offs you’re probably OK with making if you’re looking at a Framework Laptop in the first place, but it’s worth mentioning.

A few actual game benchmarks. Ones with ray-tracing features enabled tend to favor Intel’s Arc GPU, while the Radeon 890M pulls ahead in some other games.

But the new Ryzen chip’s CPU is dramatically faster than Meteor Lake at just about everything, as well as the older Ryzen 7 7840U in the older Framework board. This is the fastest the Framework Laptop has ever been, and it’s not particularly close (but if you’re waffling between the Ryzen AI version, the older AMD version that Framework sells for a bit less money or the Core Ultra 7 version, wait to see the battery life results before you spend any money). Power efficiency has also improved for heavy workloads, as demonstrated by our Handbrake video encoding tests—the Ryzen AI chip used a bit less power under heavy load and took less time to transcode our test video, so it uses quite a bit less power overall to do the same work.

Power efficiency tests under heavy load using the Handbrake transcoding tool. Test uses CPU for encoding and not hardware-accelerated GPU-assisted encoding.

We didn’t run specific performance tests on the Ryzen AI NPU, but it’s worth noting that this is also Framework’s first laptop with a neural processing unit (NPU) fast enough to support the full range of Microsoft’s Copilot+ PC features—this was one of the systems I used to test Microsoft’s near-final version of Windows Recall, for example. Intel’s other Core Ultra 100 chips, all 200-series Core Ultra chips other than the 200V series (codenamed Lunar Lake), and AMD’s Ryzen 7000- and 8000-series processors often include NPUs, but they don’t meet Microsoft’s performance requirements.

The Ryzen AI chips are also the only Copilot+ compatible processors on the market that Framework could have used while maintaining the Laptop’s current level of upgradeability. Qualcomm’s Snapdragon X Elite and Plus chips don’t support external RAM—at least, Qualcomm only lists support for soldered-down LPDDR5X in its product sheets—and Intel’s Core Ultra 200V processors use RAM integrated into the processor package itself. So if any of those features appeal to you, this is the only Framework Laptop you can buy to take advantage of them.

Battery and power

Battery tests. The Ryzen AI 300 doesn’t do great, though it’s similar to the last-gen Ryzen Framework.

When paired with the higher-resolution screen option and Framework’s 61 WHr battery, the Ryzen AI version of the laptop lasted around 8.5 hours in a PCMark Modern Office battery life test with the screen brightness set to a static 200 nits. This is a fair bit lower than the Intel Core Ultra version of the board, and it’s even worse when compared to what a MacBook Air or a more typical PC laptop will give you. But it’s holding roughly even with the older Ryzen version of the Framework board despite being much faster.

You can improve this situation somewhat by opting for the cheaper, lower-resolution screen; we didn’t test it with the Ryzen AI board, and Framework won’t sell you the lower-resolution screen with the higher-end chip. But for upgraders using the older panel, the higher-res screen reduced battery life by between 5 and 15 percent in past testing of older Framework Laptops. The slower Ryzen AI 5 and Ryzen AI 7 versions will also likely last a little longer, though Framework usually only sends us the highest-end versions of its boards to test.

A routine update

This combo screwdriver-and-spudger is still the only tool you need to take a Framework Laptop apart. Credit: Andrew Cunningham

It’s weird that my two favorite laptops right now are probably Apple’s MacBook Air and the Framework Laptop 13, but that’s where I am. They represent opposite visions of computing, each of which appeals to a different part of my brain: The MacBook Air is the personal computer at its most appliance-like, the thing you buy (or recommend) if you just don’t want to think about your computer that much. Framework embraces a more traditionally PC-like approach, favoring open standards and interoperable parts; the result is more complicated and chaotic but also more flexible. It’s the thing you buy when you like thinking about your computer.

Framework Laptop buyers continue to pay a price for getting a more repairable and modular laptop. Battery life remains OK at best, and Framework doesn’t seem to have substantially sped up its firmware or driver releases since we talked with them about it last summer. You’ll need to be comfortable taking things apart, and you’ll need to make sure you put the right expansion modules in the right bays. And you may end up paying more than you would to get the same specs from a different laptop manufacturer.

But what you get in return still feels kind of magical, and all the more so because Framework has now been shipping product for four years. The Ryzen AI version of the laptop is probably the one I’d recommend if you were buying a new one, and it’s also a huge leap forward for anyone who bought into the first-generation Framework Laptop a few years ago and is ready for an upgrade. It’s by far the fastest CPU (and, depending on the app, the fastest or second-fastest GPU) Framework has shipped in the Laptop 13. And it’s nice to at least have the option of using Copilot+ features, even if you’re not actually interested in the ones Microsoft is currently offering.

If none of the other Framework Laptops have interested you yet, this one probably won’t, either. But it’s yet another improvement in what has become a steady, consistent sequence of improvements. Mediocre battery life is hard to excuse in a laptop, but if that’s not what’s most important to you, Framework is still offering something laudable and unique.

The good

  • Framework still gets all of the basics right—a matte 3:2 LCD that’s pleasant to look at, a nice-feeling keyboard and trackpad, and a design
  • Fastest CPU ever in the Framework Laptop 13, and the fastest or second-fastest integrated GPU
  • First Framework Laptop to support Copilot+ features in Windows, if those appeal to you at all
  • Fun translucent customization options
  • Modular, upgradeable, and repairable—more so than with most laptops, you’re buying a laptop that can change along with your needs and which will be easy to refurbish or hand down to someone else when you’re ready to replace it
  • Official support for both Windows and Linux

The bad

  • Occasional glitchiness that may or may not be fixed with future firmware or driver updates
  • Some expansion modules are slower or have higher power draw if you put them in the wrong place
  • Costs more than similarly specced laptops from other OEMs
  • Still lacks certain display features some users might require or prefer—in particular, there are no OLED, touchscreen, or wide-color-gamut options

The ugly

  • Battery life remains an enduring weak point.

Photo of Andrew Cunningham

Andrew is a Senior Technology Reporter at Ars Technica, with a focus on consumer tech including computer hardware and in-depth reviews of operating systems like Windows and macOS. Andrew lives in Philadelphia and co-hosts a weekly book podcast called Overdue.

Review: Ryzen AI CPU makes this the fastest the Framework Laptop 13 has ever been Read More »

google-reveals-sky-high-gemini-usage-numbers-in-antitrust-case

Google reveals sky-high Gemini usage numbers in antitrust case

Despite the uptick in Gemini usage, Google is still far from catching OpenAI. Naturally, Google has been keeping a close eye on ChatGPT traffic. OpenAI has also seen traffic increase, putting ChatGPT around 600 million monthly active users, according to Google’s analysis. Early this year, reports pegged ChatGPT usage at around 400 million users per month.

There are many ways to measure web traffic, and not all of them tell you what you might think. For example, OpenAI has recently claimed weekly traffic as high as 400 million, but companies can choose the seven-day period in a given month they report as weekly active users. A monthly metric is more straightforward, and we have some degree of trust that Google isn’t using fake or unreliable numbers in a case where the company’s past conduct has already harmed its legal position.

While all AI firms strive to lock in as many users as possible, this is not the total win it would be for a retail site or social media platform—each person using Gemini or ChatGPT costs the company money because generative AI is so computationally expensive. Google doesn’t talk about how much it earns (more likely loses) from Gemini subscriptions, but OpenAI has noted that it loses money even on its $200 monthly plan. So while having a broad user base is essential to make these products viable in the long term, it just means higher costs unless the cost of running massive AI models comes down.

Google reveals sky-high Gemini usage numbers in antitrust case Read More »

openai-wants-to-buy-chrome-and-make-it-an-“ai-first”-experience

OpenAI wants to buy Chrome and make it an “AI-first” experience

According to Turley, OpenAI would throw its proverbial hat in the ring if Google had to sell. When asked if OpenAI would want Chrome, he was unequivocal. “Yes, we would, as would many other parties,” Turley said.

OpenAI has reportedly considered building its own Chromium-based browser to compete with Chrome. Several months ago, the company hired former Google developers Ben Goodger and Darin Fisher, both of whom worked to bring Chrome to market.

Close-up of Google Chrome Web Browser web page on the web browser. Chrome is widely used web browser developed by Google.

Credit: Getty Images

It’s not hard to see why OpenAI might want a browser, particularly Chrome with its 4 billion users and 67 percent market share. Chrome would instantly give OpenAI a massive install base of users who have been incentivized to use Google services. If OpenAI were running the show, you can bet ChatGPT would be integrated throughout the experience—Turley said as much, predicting an “AI-first” experience. The user data flowing to the owner of Chrome could also be invaluable in training agentic AI models that can operate browsers on the user’s behalf.

Interestingly, there’s so much discussion about who should buy Chrome, but relatively little about spinning off Chrome into an independent company. Google has contended that Chrome can’t survive on its own. However, the existence of Google’s multibillion-dollar search placement deals, which the DOJ wants to end, suggests otherwise. Regardless, if Google has to sell, and OpenAI has the cash, we might get the proposed “AI-first” browsing experience.

OpenAI wants to buy Chrome and make it an “AI-first” experience Read More »

google-won’t-ditch-third-party-cookies-in-chrome-after-all

Google won’t ditch third-party cookies in Chrome after all

Maintaining the status quo

While Google’s sandbox project is looking more directionless today, it is not completely ending the initiative. The team still plans to deploy promised improvements in Chrome’s Incognito Mode, which has been re-architected to preserve user privacy after numerous complaints. Incognito Mode blocks all third-party cookies, and later this year, it will gain IP protection, which masks a user’s IP address to protect against cross-site tracking.

What is Topics?

Chavez admits that this change will mean Google’s Privacy Sandbox APIs will have a “different role to play” in the market. That’s a kind way to put it. Google will continue developing these tools and will work with industry partners to find a path forward in the coming months. The company still hopes to see adoption of the Privacy Sandbox increase, but the industry is unlikely to give up on cookies voluntarily.

While Google focuses on how ad privacy has improved since it began working on the Privacy Sandbox, the changes in Google’s legal exposure are probably more relevant. Since launching the program, Google has lost three antitrust cases, two of which are relevant here: the search case currently in the remedy phase and the newly decided ad tech case. As the government begins arguing that Chrome gives Google too much power, it would be a bad look to force a realignment of the advertising industry using the dominance of Chrome.

In some ways, this is a loss—tracking cookies are undeniably terrible, and Google’s proposed alternative is better for privacy, at least on paper. However, universal adoption of the Privacy Sandbox could also give Google more power than it already has, and the supposed privacy advantages may never have fully materialized as Google continues to seek higher revenue.

Google won’t ditch third-party cookies in Chrome after all Read More »

man-buys-racetrack,-ends-up-launching-the-netflix-of-grassroots-motorsports

Man buys racetrack, ends up launching the Netflix of grassroots motorsports


FRDM+ is profitable, has its own smart TV apps. Subscriptions start at $20/month.

In 2019, Garrett Mitchell was already an Internet success. His YouTube channel, Cleetus McFarland, had over a million followers. If you perused the channel at that time, you would’ve found a range of grassroots motorsports videos with the type of vehicular shenanigans that earn truckloads of views. Some of those older videos include “BLEW BY A COP AT 120+mph! OOPS!,” “THERE’S A T-REX ON THE TRACK!,” and “Manual Transmission With Paddle Shifters!?!.”

Those videos made Mitchell, aka Cleetus McFarland, a known personality among automotive enthusiasts. But the YouTuber wanted more financial independence beyond the Google platform and firms willing to sponsor his channel.

“… after my YouTube was growing and some of my antics [were] getting videos de-monetized, I realized I needed a playground,” Mitchell told Ars Technica in an email.

Mitchell found a road toward new monetization opportunities through the DeSoto Super Speedway. The Bradenton, Florida, track had changed ownership multiple times since opening in the 1970s. The oval-shaped racetrack is three-eighths of a mile long with 12-degree banking angles.

BRADENTON, FL — Mid-1980s: Late Model racing action at DeSoto Speedway in the mid-1980s. Both the All-Pro Series and NASCAR All-American Challenge Series ran races at the track in 1985 and 1986.

BRADENTON, FL — Mid-1980s: Late Model racing action at DeSoto Speedway in the mid-1980s. Both the All-Pro Series and NASCAR All-American Challenge Series ran races at the track in 1985 and 1986. Credit: ISC Images & Archives via Getty Images

By 2018, the track had closed its doors and was going unused. DeSoto happened to be next to Mitchell’s favorite drag strip, giving the YouTuber the idea of turning it into a stadium where people could watch burnouts and other “massive, rowdy” ticketed events. Mitchell added:

So I sold everything I could, borrowed some money from my business manager, and went all in for [$]2.2 million.

But like the rest of the world, Mitchell hit the brakes on his 2020 plans during COVID-19 lockdowns. Soon after his purchase, Mitchell couldn’t use the track, renamed Freedom Factory, for large gatherings, forcing him to reconsider his plans.

“We had no other option but to entertain the people somehow. And with no other racing goin’ on anywhere, we bet big on making something happen. And it worked,” Mitchell said.

That “something” was a pay-per-view (PPV) event hosted from the Freedom Factory in April 2020. The event led to others and, eventually, Mitchell running his own subscription video on demand (SVOD) service, FRDM+, which originally launched as Cleetervision in 2022.

Today, a FRDM+ subscription costs $20 per month or $120 per year. A subscription provides access to an impressive library of automotive videos. Some are archived from Mitchell’s YouTube channel. Other, exclusive videos feature content such as interviews with motorsports influencers and members of Mitchell’s staff and crew, and outrageous motorsports stunts. You can watch videos from other influencers on FRDM+, and the business can also white-label its platform into other influencers’ websites, too.

“A race against time”

Before Mitchell could host his first PPV event, he had to prepare the speedway. Explaining the ordeal to Ars, he wrote:

We cleaned that place up best we could, but let’s be real, it was rough. Lights were out, weeds poppin’ up through the asphalt, the whole deal.

Pulling off the first PPV event at the Freedom Factory speedway was a “race against time,” Jonny Mill, who built FRDM+’s tech stack and serves as company president, told Ars.

“Florida implemented a statewide shutdown on the very day of our event,” he said.

Mitchell also struggled to get the right workers and equipment needed for the PPV. Flights weren’t available due to the pandemic, forcing Mill to produce the event from California using a cell phone group chat and “last-minute local crew,” per Mitchell. The ENG camera person was much shorter than Mitchell “and had to climb on whatever she could just to keep me in frame,” he recalled.

Mitchell said Freedom Factory’s first PPV event had 75,000 concurrent viewers, which caused his website and those of the event sponsors to crash.

“Our initial bandwidth provider laughed at our viewership projections, and, of course, we surpassed them in the first week of pre-sales,” Mill said. “They did apologize before asking for a much larger check.”

Other early obstacles included determining how to embed the livestream platform into Mitchell’s e-commerce site. The biggest challenge there was “juggling two separate logins, one for merch shopping and another for livestream PPV, all within the same site,” Mill explained.

“Now, our focus is on seamlessly guiding the YouTube audience over to FRDM+ for premium live events,” he added.

Live events are still the heart of FRDM+. The service had 21 livestreamed events scheduled throughout 2025, and more are expected to come.

Peeking under the hood

Today, bandwidth isn’t a problem for FRDM+, and navigating the streaming service doesn’t feel much different from something like Netflix. There are different “channels” (grouped together by related content or ongoing series) on top and new releases and upcoming content highlighted below. There are horizontal scrolling rows, and many titles have content summaries and/or trailers. The platform also has a support section with instructions for canceling subscriptions.

A screenshot of FRDM+

Browsing FRDM+.

Browsing FRDM+. Credit: FRDM+

Like with other SVOD services, subscribers can watch FRDM+ via a web browser or through a smart TV app. FRDM+ currently has apps for Apple TV, Fire OS, and Roku OS. Mitchell said the team’s constantly working on more connected TV apps, as well as adding features, “more interactivity,” and customers.

To keep the wheels spinning, FRDM+ leverages a diverse range of technologies, Mill explained:

At the core of our infrastructure, AWS bandwidth servers handle the heavy lifting, while Accedo powers the connected TV apps, bridging the gap between our tech stack and the audience. Brightcove serves as our primary video player partner, with additional backup systems in place to maintain reliability.

For a service like this, with live events, redundancy is critical, Mill said.

“At the Freedom Factory, we even beam air fiber from a house five miles away to ensure a reliable second Internet. We also have a hidden page on [the Cleetus McFarland website] to launch a backup stream if the primary one fails,” he said.

Today, FRDM+’s biggest challenge isn’t a technical one. Instead, it’s around managing the business’s different parts using a small team. FRDM+ has 35 full-time employees across its Shop, Race Track, Events, and Merch divisions and is “entirely self-funded,” per Mill. The company also relies on contractors for productions, but its core livestream team has six full-time employees.

Mitchell told Ars that FRDM+ is profitable, but he couldn’t get into specifics. He said the service has “strong year-over-year growth and a solid financial foundation that allows us to continue reinvesting in our team and services,” like a “robust technology stack, larger events, venue rentals, and even giving away helicopters and Lamborghinis as the prizes for our races.”

“Having been at Discovery during the launch of MotorTrend OnDemand, I’ve witnessed the power of substantial budgets firsthand,” Mill said. “Yet, [FRDM+ has] achieved greater success organically than [Discovery] did with their eight-figure marketing investment. This autonomy and efficiency are a testament to the strength of our approach.”

Any profitability for a 3-year-old streaming service is commendable. Due to wildly differing audiences, markets, costs, and scales, comparing FRDM+’s financials to the likes of Netflix and other mainstream streaming services is like comparing apples to oranges. But it’s interesting to consider that FRDM+ has achieved profitability faster than some of those services, like Peacock, which also launched in 2020, and Apple TV+, which debuted in 2019.

FRDM+ doesn’t share subscription numbers publicly, but Mitchell told Ars that the subscription service has a 93 percent retention. Mill attributed that number to a loyal, engaged community driven by direct communication with Mitchell.

Mill also suggested to Ars that FRDM+ has successfully converted over 5 percent of Mitchell’s YouTube audience. Five percent of Cleetus McFarland’s current YouTube base would be 212,500 people.

Photo of Scharon Harding

Scharon is a Senior Technology Reporter at Ars Technica writing news, reviews, and analysis on consumer gadgets and services. She’s been reporting on technology for over 10 years, with bylines at Tom’s Hardware, Channelnomics, and CRN UK.

Man buys racetrack, ends up launching the Netflix of grassroots motorsports Read More »

google-messages-can-now-blur-unwanted-nudes,-remind-people-not-to-send-them

Google Messages can now blur unwanted nudes, remind people not to send them

Google announced last year that it would deploy safety tools in Google Messages to help users avoid unwanted nudes by automatically blurring the content. Now, that feature is finally beginning to roll out. Spicy image-blurring may be enabled by default on some devices, but others will need to turn it on manually. If you don’t see the option yet, don’t fret. Sensitive Content Warnings will arrive on most of the world’s Android phones soon enough.

If you’re an adult using an unrestricted phone, Sensitive Content Warnings will be disabled by default. For teenagers using unsupervised phones, the feature is enabled but can be disabled in the Messages settings. On supervised kids’ phones, the feature is enabled and cannot be disabled on-device. Only the Family Link administrator can do that. For everyone else, the settings are available in the Messages app settings under Protection and Safety.

To make the feature sufficiently private, all the detection happens on the device. As a result, there was some consternation among Android users when the necessary components began rolling out over the last few months. For people who carefully control the software installed on their mobile devices, the sudden appearance of a package called SafetyCore was an affront to the sanctity of their phones. While you can remove the app (it’s listed under “Android System SafetyCore”), it doesn’t take up much space and won’t be active unless you enable Sensitive Content Warnings.

Google Messages can now blur unwanted nudes, remind people not to send them Read More »

in-depth-with-windows-11-recall—and-what-microsoft-has-(and-hasn’t)-fixed

In depth with Windows 11 Recall—and what Microsoft has (and hasn’t) fixed


Original botched launch still haunts new version of data-scraping AI feature.

Recall is coming back. Credit: Andrew Cunningham

Recall is coming back. Credit: Andrew Cunningham

Microsoft is preparing to reintroduce Recall to Windows 11. A feature limited to Copilot+ PCs—a label that just a fraction of a fraction of Windows 11 systems even qualify for—Recall has been controversial in part because it builds an extensive database of text and screenshots that records almost everything you do on your PC.

But the main problem with the initial version of Recall—the one that was delayed at the last minute after a large-scale outcry from security researchers, reporters, and users—was not just that it recorded everything you did on your PC but that it was a rushed, enabled-by-default feature with gaping security holes that made it trivial for anyone with any kind of access to your PC to see your entire Recall database.

It made no efforts to automatically exclude sensitive data like bank information or credit card numbers, offering just a few mechanisms to users to manually exclude specific apps or websites. It had been built quickly, outside of the normal extensive Windows Insider preview and testing process. And all of this was happening at the same time that the company was pledging to prioritize security over all other considerations, following several serious and highly public breaches.

Any coverage of the current version of Recall should mention what has changed since then.

Recall is being rolled out to Microsoft’s Windows Insider Release Preview channel after months of testing in the more experimental and less-stable channels, just like most other Windows features. It’s turned off by default and can be removed from Windows root-and-branch by users and IT administrators who don’t want it there. Microsoft has overhauled the feature’s underlying security architecture, encrypting data at rest so it can’t be accessed by other users on the PC, adding automated filters to screen out sensitive information, and requiring frequent reauthentication with Windows Hello anytime a user accesses their own Recall database.

Testing how Recall works

I installed the Release Preview Windows 11 build with Recall on a Snapdragon X Elite version of the Surface Laptop and a couple of Ryzen AI PCs, which all have NPUs fast enough to support the Copilot+ features.

No Windows PCs without this NPU will offer Recall or any other Copilot+ features—that’s every single PC sold before mid-2024 and the vast majority of PCs since then. Users may come up with ways to run those features on unsupported hardware some other way. But by default, Recall isn’t something most of Windows’ current user base will have to worry about.

Microsoft is taking data protection more seriously this time around. If Windows Hello isn’t enabled or drive encryption isn’t turned on, Recall will refuse to start working until you fix the issues. Credit: Andrew Cunningham

After installing the update, you’ll see a single OOBE-style setup screen describing Recall and offering to turn it on; as promised, it is now off by default until you opt in. And even if you accept Recall on this screen, you have to opt in a second time as part of the Recall setup to actually turn the feature on. We’ll be on high alert for a bait-and-switch when Microsoft is ready to remove Recall’s “preview” label, whenever that happens, but at least for now, opt-in means opt-in.

Enable Recall, and the snapshotting begins. As before, it’s storing two things: actual screenshots of the active area of your screen, minus the taskbar, and a searchable database of text that it scrapes from those screenshots using OCR. Somewhat oddly, there are limits on what Recall will offer to OCR for you; even if you’re using multiple apps onscreen at the same time, only the active, currently-in-focus app seems to have its text scraped and stored.

This is also more or less how Recall handles multi-monitor support; only the active display has screenshots taken, and only the active window on the active display is OCR’d. This does prevent Recall from taking gigabytes and gigabytes of screenshots of static or empty monitors, though it means the app may miss capturing content that updates passively if you don’t interact with those windows periodically.

All of this OCR’d text is fully searchable and can be copied directly from Recall to be pasted somewhere else. Recall will also offer to open whatever app or website is visible in the screenshot, and it gives you the option to delete that specific screenshot and all screenshots from specific apps (handy, if you decide you want to add an entire app to your filtering settings and you want to get rid of all existing snapshots of it).

Here are some basic facts about how Recall works on a PC since there’s a lot of FUD circulating about this, and much of the information on the Internet is about the older, insecure version from last year:

  • Recall is per-user. Setting up Recall for one user account does not turn on Recall for all users of a PC.
  • Recall does not require a Microsoft account.
  • Recall does not require an Internet connection or any cloud-side processing to work.
  • Recall does require your local disk to be encrypted with Device Encryption/BitLocker.
  • Recall does require Windows Hello and either a fingerprint reader or face-scanning camera for setup, though once it’s set up, it can be unlocked with a Windows Hello PIN.
  • Windows Hello authentication happens every time you open the Recall app.
  • Enabling Recall and changing its settings does not require an administrator account.
  • Recall can be uninstalled entirely by unchecking it in the legacy Windows Features control panel (you can also search for “turn Windows features on and off”).

If you read our coverage of the initial version, there’s a whole lot about how Recall functions that’s essentially the same as it was before. In Settings, you can see how much storage the feature is using and limit the total amount of storage Recall can use. The amount of time a snapshot can be kept is normally determined by the amount of space available, not by the age of the snapshot, but you can optionally choose a second age-based expiration date for snapshots (options range from 30 to 180 days).

You can see Recall hit the system’s NPU periodically every time it takes a snapshot (this is on an AMD Ryzen AI system, but it should be the same for Qualcomm Snapdragon PCs and Intel Core Ultra/Lunar Lake systems). Browsing your Recall database doesn’t use the NPU. Credit: Andrew Cunningham

It’s also possible to delete the entire database or all recent snapshots (those from the past hour, past day, past week, or past month), toggle the automated filtering of sensitive content, or add specific apps and websites you’d like to have filtered. Recall can temporarily be paused by clicking the system tray icon (which is always visible when you have Recall turned on), and it can be turned off entirely in Settings. Neither of these options will delete existing snapshots; they just stop your PC from creating new ones.

The amount of space Recall needs to do its thing will depend on a bunch of factors, including how actively you use your PC and how many things you filter out. But in my experience, it can easily generate a couple of hundred megabytes per day of images. A Ryzen system with a 1TB SSD allocated 150GB of space to Recall snapshots by default, but even a smaller 25GB Recall database could easily store a few months of data.

Fixes: Improved filtering, encryption at rest

For apps and sites that you know you don’t want to end up in Recall, you can manually add them to the exclusion lists in the Settings app. As a rule, major browsers running in private or incognito modes are also generally not snapshotted.

If you have an app that’s being filtered onscreen for any reason—even if it’s onscreen at the same time as an app that’s not being filtered, Recall won’t take pictures of your desktop at all. I ran an InPrivate Microsoft Edge window next to a regular window, and Microsoft’s solution is just to avoid capturing and storing screenshots entirely rather than filtering or blanking out the filtered app or site in some way.

This is probably the best way to do it! It minimizes the risk of anything being captured accidentally just because it’s running in the background, for example. But it could mean you don’t end up capturing much in Recall at all if you’re frequently mixing filtered and unfiltered apps.

New to this version of Recall is an attempt at automated content filtering to address one of the major concerns about the original iteration of Recall—that it can capture and store sensitive information like credit card numbers and passwords. This filtering is based on the technology Microsoft uses for Microsoft Purview Information Protection, an enterprise feature used to tag sensitive information on business, healthcare, and government systems.

This automated content filtering is hit and miss. Recall wouldn’t take snapshots of a webpage with a visible credit card field, or my online banking site, or an image of my driver’s license, or a recent pay stub, or of the Bitwarden password manager while viewing credentials. But I managed to find edge cases in less than five minutes, and you’ll be able to find them, too; Recall saved snapshots showing a recent check, with the account holder’s name, address, and account and routing numbers visible, and others testing it have still caught it recording credit card information in some cases.

The automated filtering is still a big improvement from before, when it would capture this kind of information indiscriminately. But things will inevitably slip through, and the automated filtering won’t help at all with other kinds of data; Recall will take pictures of email and messaging apps without distinguishing between what’s sensitive (school information for my kid, emails about Microsoft’s own product embargoes) and what isn’t.

Recall can be removed entirely. If you take it out, it’s totally gone—the options to configure it won’t even appear in Settings anymore. Credit: Andrew Cunningham

The upshot is that if you capture months and months and gigabytes and gigabytes of Recall data on your PC, it’s inevitable that it will capture something you probably wouldn’t want to be preserved in an easily searchable database.

One issue is that there’s no easy way to check and confirm what Recall is and isn’t filtering without actually scrolling through the database and checking snapshots manually. The system tray status icon does change to display a small triangle and will show you a “some content is being filtered” status message when something is being filtered, but the system won’t tell you what it is; I have some kind of filtered app or browser tab open somewhere right now, and I have no idea which one it is because Windows won’t tell me. That any attempt at automated filtering is hit-and-miss should be expected, but more transparency would help instill trust and help users fine-tune their filtering settings.

Recall’s files are still clearly visible and trivial to access, but with one improvement: They’re all actually encrypted now. Credit: Andrew Cunningham

Microsoft also seems to have fixed the single largest problem with Recall: previously, all screenshots and the entire text database were stored in plaintext with zero encryption. It was technicallyusually encrypted, insofar as the entire SSD in a modern PC is encrypted when you sign into a Microsoft account or enable Bitlocker, but any user with any kind of access to your PC (either physical or remote) could easily grab those files and view them anywhere with no additional authentication necessary.

This is fixed now. Recall’s entire file structure is available for anyone to look at, stored away in the user’s AppData folder in a directory called CoreAIPlatform.00UKP. Other administrators on the same PC can still navigate to these folders from a different user account and move or copy the files. Encryption renders them (hypothetically) unreadable.

Microsoft has gone into some detail about exactly how it’s protecting and storing the encryption keys used to encrypt these files—the company says “all encryption keys [are] protected by a hypervisor or TPM.” Rate-limiting and “anti-hammering” protections are also in place to protect Recall data, though I kind of have to take Microsoft at its word on that one.

That said, I don’t love that it’s still possible to get at those files at all. It leaves open the possibility that someone could theoretically grab a few megabytes’ worth of data. But it’s now much harder to get at that data, and better filtering means what is in there should be slightly less all-encompassing.

Lingering technical issues

As we mentioned already, Microsoft’s automated content filtering is hit-and-miss. Certainly, there’s a lot of stuff that the original version of Recall would capture that the new one won’t, but I didn’t have to work hard to find corner-cases, and you probably won’t, either. Turning Recall on still means assuming risk and being comfortable with the data and authentication protections Microsoft has implemented.

We’d also like there to be a way for apps to tell Recall to exclude them by default, which would be useful for password managers, encrypted messaging apps, and any other software where privacy is meant to be the point. Yes, users can choose to exclude these apps from Recall backups themselves. But as with Recall itself, opting in to having that data collected would be preferable to needing to opt out.

You need a fingerprint reader or face-scanning camera to get Recall set up, but once it is set up, anyone with your PIN and access to your PC can get in and see all your stuff. Credit: Andrew Cunningham

Another issue is that, while Recall does require a fingerprint reader or face-scanning camera when you set it up the very first time, you can unlock it with a Windows Hello PIN after it’s already going.

Microsoft has said that this is meant to be a fallback option in case you need to access your Recall database and there’s some kind of hardware issue with your fingerprint sensor. But in practice, it feels like too easy a workaround for a domestic abuser or someone else with access to your PC and a reason to know your PIN (and note that the PIN also gets them into your PC in the first place, so encryption isn’t really a fix for this). It feels like too broad a solution for a relatively rare problem.

Security researcher Kevin Beaumont, whose testing helped call attention to the problems with the original version of Recall last year, identified this as one of Recall’s biggest outstanding technical problems in a blog post shared with Ars Technica shortly before its publication (as of this writing, it’s available here; he and I also exchanged multiple text over the weekend comparing our findings).

“In my opinion, requiring devices to have enhanced biometrics with Windows Hello  but then not requiring said biometrics to actually access Recall snapshots is a big problem,” Beaumont wrote. “It will create a false sense of security in customers and false downstream advertising about the security of Recall.”

Beaumont also noted that, while the encryption on the Recall snapshots and database made it a “much, much better design,” “all hell would break loose” if attackers ever worked out a way to bypass this encryption.

“Microsoft know this and have invested in trying to stop it by encrypting the database files, but given I live in the trenches where ransomware groups are running around with zero days in Windows on an almost monthly basis nowadays, where patches arrive months later… Lord, this could go wrong,” he wrote.

But most of what’s wrong with Recall is harder to fix

Microsoft has actually addressed many of the specific, substantive Recall complaints raised by security researchers and our own reporting. It’s gone through the standard Windows testing process and has been available in public preview in its current form since late November. And yet the knee-jerk reaction to Recall news is still generally to treat it as though it were the same botched, bug-riddled software that nearly shipped last summer.

Some of this is the asymmetrical nature of how news spreads on the Internet—without revealing traffic data, I’ll just say that articles about Recall having problems have been read many, many more times by many more people than pieces about the steps Microsoft has taken to fix Recall. The latter reports simply aren’t being encountered by many of the minds Microsoft needs to change.

But the other problem goes deeper than the technology itself and gets back to something I brought up in my first Recall preview nearly a year ago—regardless of how it is architected and regardless of how many privacy policies and reassurances the company publishes, people simply don’t trust Microsoft enough to be excited about “the feature that records and stores every single thing you do with your PC.”

Recall continues to demand an extraordinary level of trust that Microsoft hasn’t earned. However secure and private it is—and, again, the version people will actually get is much better than the version that caused the original controversy—it just feels creepy to open up the app and see confidential work materials and pictures of your kid. You’re already trusting Microsoft with those things any time you use your PC, but there’s something viscerally unsettling about actually seeing evidence that your computer is tracking you, even if you’re not doing anything you’re worried about hiding, even if you’ve excluded certain apps or sites, and even if you “know” that part of the reason why Recall requires a Copilot+ PC is because it’s processing everything locally rather than on a server somewhere.

This was a problem that Microsoft made exponentially worse by screwing up the Recall rollout so badly in the first place. Recall made the kind of ugly first impression that it’s hard to dig out from under, no matter how thoroughly you fix the underlying problems. It’s Windows Vista. It’s Apple Maps. It’s the Android tablet.

And in doing that kind of damage to Recall (and possibly also to the broader Copilot+ branding project), Microsoft has practically guaranteed that many users will refuse to turn it on or uninstall it entirely, no matter how it actually works or how well the initial problems have been addressed.

Unfortunately, those people probably have it right. I can see no signs that Recall data is as easily accessed or compromised as before or that Microsoft is sending any Recall data from my PC to anywhere else. But today’s Microsoft has earned itself distrust-by-default from many users, thanks not just to the sloppy Recall rollout but also to the endless ads and aggressive cross-promotion of its own products that dominate modern Windows versions. That’s the kind of problem you can’t patch your way out of.

Listing image: Andrew Cunningham

Photo of Andrew Cunningham

Andrew is a Senior Technology Reporter at Ars Technica, with a focus on consumer tech including computer hardware and in-depth reviews of operating systems like Windows and macOS. Andrew lives in Philadelphia and co-hosts a weekly book podcast called Overdue.

In depth with Windows 11 Recall—and what Microsoft has (and hasn’t) fixed Read More »