from the the-law-breaking-admin dept

We’ve been pointing out the fundamental contradiction at the heart of mandatory age verification laws for years now. To verify someone’s age online, you have to collect personal data from them. If that someone turns out to be a child, congratulations: you’ve just collected personal data from a child without parental consent. Which is a direct violation of the Children’s Online Privacy Protection Act (COPPA)—the very law that’s supposed to be protecting kids.

So what happens when the agency charged with enforcing COPPA finally notices this obvious problem? If you guessed “they admit the conflict and then just promise not to enforce the law,” you’d be exactly right.

The FTC put out a policy statement last week that is remarkable in what it tacitly concedes:

The Federal Trade Commission issued a policy statement today announcing that the Commission will not bring an enforcement action under the Children’s Online Privacy Protection Rule (COPPA Rule) against certain website and online service operators that collect, use, and disclose personal information for the sole purpose of determining a user’s age via age verification technologies.

The FTC appears to be explicitly acknowledging that age verification technologies involve collecting personal information from users—including children—in a way that would otherwise trigger COPPA liability. If the technology didn’t create a COPPA problem, there would be no need for a policy statement promising non-enforcement. You don’t issue a formal announcement saying “we won’t sue you for this” unless “this” is something you could, in fact, sue people for.

The statement itself tries to dress this up by noting that age verification tech “may require the collection of personal information from children, prompting questions about whether such activities could violate the COPPA Rule.” But “prompting questions” is doing an awful lot of work in that sentence. The answer to those questions is pretty obviously “yes, collecting personal information from children without parental consent violates the rule that says you can’t collect personal information from children without parental consent.” The FTC just doesn’t want to say that part out loud, because then the follow-up question becomes: “so why are you encouraging companies to do it?”

Instead, they’ve decided to create an enforcement carve-out. Do the thing that violates the law, but pinky-promise you’ll only use the data to check the kid’s age, delete it afterward, and keep it secure. Then we won’t come after you. This is the FTC solving a legal contradiction not by asking Congress to fix the underlying law or admitting the technology is fundamentally flawed, but by deciding to selectively not enforce the law it’s supposed to be enforcing.

The honest approach would have been to tell Congress that age verification, as currently conceived, cannot be squared with existing privacy law—and that if lawmakers want it anyway, they need to resolve that conflict themselves rather than asking the FTC to pretend it doesn’t exist.

No such luck.

And boy, do they seem proud of themselves. Here’s Christopher Mufarrige, Director of the FTC’s Bureau of Consumer Protection:

“Age verification technologies are some of the most child-protective technologies to emerge in decades…. Our statement incentivizes operators to use these innovative tools, empowering parents to protect their children online.”

“The most child-protective technologies to emerge in decades.”

Excuse me, what?

This is the kind of statement that sounds authoritative right up until you spend thirty seconds thinking about it. Anyone with any knowledge of security and privacy knows that age verification is anything but “child protective.” It involves a huge invasion of privacy, for extremely faulty technology, that has all sorts of downstream effects that put kids at risk.

Oh, and the FTC seems proud that the vote for this was unanimous—though it’s worth noting that Donald Trump fired the two Democratic members of the FTC and has made no apparent efforts to replace them, despite Congress designating that the FTC is supposed to have five full members, with two from the opposing party. A unanimous vote among the remaining two Republicans is a strange thing to brag about.

The FTC even posted about this on X, and the response was… well, let me just show you:

If you can’t see that, the main part to pay attention to is not the tweet from the FTC itself, but the Community Note that (under the way Community Notes works, notes need widespread consensus among users to be appended to the public tweet):

Readers added context they thought people might want to know

Contrary to their claim, using age verification has numerous issues, including but not limited to:

1. Easily bypassed

Advertisement

2. Risks of security data breach

3. Inaccuracies (Placing adults into underage groups, vice versa)

And many more… (sigh, I need a break).

Yeah, we all need a break.

That Community Note does a better job explaining the state of age verification technology than the FTC’s entire Bureau of Consumer Protection. It methodically lists out the problems: kids easily bypass these systems, the collected data creates massive security breach risks, and the technology produces wildly inaccurate results that lock adults out while letting kids through (and vice versa). When the consensus-driven crowdsourced fact-check on your own announcement is more informative than the announcement itself, maybe it’s time to reconsider the announcement.

But let’s say, for the sake of argument, that the technology worked perfectly. Would mandatory age verification still be a good idea?

That still wouldn’t solve the issues with this technology and the harm it does to kids. Even UNICEF (UNICEF!) has been warning that age restriction approaches can actively harm the children they’re supposed to protect. After Australia’s social media ban for under-16s went into effect, UNICEF put out a statement that could not have been more clear about the risks:

“While UNICEF welcomes the growing commitment to children’s online safety, social media bans come with their own risks, and they may even backfire,” the agency said in a statement.

For many children, particularly those who are isolated or marginalised, social media is a lifeline for learning, connection, play and self-expression, UNICEF explained.

Advertisement

Moreover, many will still access social media – for example, through workarounds, shared devices, or use of less regulated platforms – which will only make it harder to protect them.

So the actual child welfare experts are saying that age verification can backfire, push kids into less safe spaces, and should never be treated as a substitute for real safety measures. Meanwhile, the FTC is calling the same technology “the most child-protective” thing to come along in a generation and is waiving its own enforcement authority to encourage more of it.

What we have here is a federal agency that has identified a direct conflict between the law it enforces and the policy outcome it wants. Rather than grappling with what that conflict means—maybe age verification as currently conceived just doesn’t work within the existing legal framework, and for good reason—the FTC has chosen to simply look the other way. The message to companies is clear: go ahead and collect data from kids to figure out if they’re kids. We know that violates COPPA. We don’t care. We like age verification more than we like enforcing our own rules.

That’s a hell of a policy position for the agency that’s supposed to be the last line of defense for children’s privacy online.

Filed Under: age verification, coppa, ftc, think of the children

Several key tech companies signed a nonbinding pledge at the White House on Wednesday that the Trump administration claims will ensure that tech companies do not pass the cost of data centers on to consumers’ utility bills.

“Data centers … they need some PR help,” President Donald Trump said at the event. “People think that if the data center goes in, their electricity is going to go up.”

He was flanked by representatives from Microsoft, Meta, OpenAI, xAI, Google/Alphabet, Oracle, and Amazon.

Bipartisan anger about data centers and their potential impact on consumers’ electric bills has exploded over the past year. As the White House goes all in on AI, the pledge marks a significant salvo by the Trump administration to assure voters that they will not be affected by rising costs.

But electricity experts and industry insiders threw doubt on how much power the White House actually has to create meaningful consumer protections.

“This is theater,” says Ari Peskoe, the director of the Electricity Law Initiative at the Harvard Law School Environmental and Energy Law Program. “This is a press release designed to make it seem like they are addressing this issue. But this issue can only really be addressed by utility regulators or Congress. The White House doesn’t really have a lot of moves here, and I don’t think the tech companies themselves are the most important parties on cost issues.”

The White House did not immediately respond to a request for comment.

Data centers played a key role in last year’s elections in certain states, including Georgia and Virginia, and are factoring into other races playing out across the country this month. A recent poll conducted by Heatmap News shows that fewer than 30 percent of American voters would support a data center being built near where they live. A number of states have introduced moratoriums on data centers into their state legislatures this year, while others have bills that would seek to help offload the cost from the consumer to the companies building and operating the facilities.

Over the past few months, some big tech companies—including Microsoft and Anthropic—have rolled out various pledges around their data center construction and operation. These pledges follow multiple reports that the president was seeking assurances from tech companies to help take the costs of data centers off American consumers.

In late January, Trump wrote in a Truth Social post that Democrats were to blame for high electricity costs and that he was “working with major American Technology Companies” to ensure “Americans don’t ‘pick up the tab’ for their POWER consumption, in the form of paying higher Utility bills.” Less than a month later, he said during his State of the Union address that he would introduce a “ratepayer protection pledge.”

“We’re telling the major tech companies that they have the obligation to provide for their own power needs,” he said. “They can build their own power plants as part of their factory, so that no one’s prices will go up and, in many cases, prices of electricity will go down for the community, and very substantially then.”

The pledges made independently by key tech companies this year, and the one signed Wednesday, reiterate a lot of promises and initiatives that some tech companies have already been working on. In a blog post published by Google highlighting its commitment to the pledge, the company lists several ongoing initiatives, including investments in nuclear and geothermal energy as well as agreement frameworks with electric utilities and pledges to invest in job creation.

Disposable vapes aren’t quite the problem/resource stream they once were, with many jurisdictions moving to ban the absurdly wasteful little devices, but there are still a lot of slightly-smelly lithium batteries in the wild. You might be forgiven for thinking that most of them seem to be in [Chris Doel]’s UK workshop, given that he’s now cruising around what has to be the world’s only vape-powered car.

Technically, anyway; some motorheads might object to calling donor vehicle [Chris] starts with a car, but the venerable G-Wiz has four wheels, four seats, lights and a windscreen, so what more do you want? Horsepower in excess of 17 ponies (12.6 kW)? Top speeds in excess of 50 Mph (80 km/h)? Something other than the dead weight of 20-year-old lead-acid batteries? Well, [Chris] at least fixes that last part.

The conversion is amazingly simple: he just straps his 500 disposable vape battery pack into the back seat– the same one that was powering his shop–into the GWiz, and it’s off to the races. Not quickly, mind you, but with 500 lightly-used lithium cells in the back seat, how fast would you want to go? Hopefully the power bank goes back on the wall after the test drive, or he finds a better mounting solution. To [Chris]’s credit, he did renovate his pack with extra support and insulation, and put all the cells in an insulated aluminum box. Still, the low speed has to count as a safety feature at this point.

Charging isn’t fast either, as [Chris] has made the probably-controversial decision to use USB-C. We usually approve of USB-Cing all the things, but a car might be taking things too far, even one with such a comparatively tiny battery. Perhaps his earlier (equally nicotine-soaked) e-bike project would have been a better fit for USB charging.

Thanks to [Vaughna] for the tip!

Educators have seen wave after wave of “innovative” solutions promise to address long-standing challenges — from personalization and engagement to college- and career-readiness — yet many issues remain stubbornly unresolved. Too often, solutions are developed and scaled without a clear understanding of how challenges show up in daily classroom experiences or how students, families and educators define the problems.

Understanding the everyday barriers that students, families, practitioners and administrators identify ensures that potential solutions — whether technological, instructional or relational — are grounded in real needs rather than assumptions.

What These Challenges Look Like in Classrooms and Systems

In Digital Promise’s co-research and co-design work with communities across the country, students and educators describe challenges that are neither new nor isolated, but reflect enduring gaps in how learning environments are designed and supported. Looking closely at how these challenges surface through our Challenge Map reveals the deep connections between instructional practice, student engagement and systems-level supports — and why tackling one without the others often falls short.

Together, these experiences shape whether students feel their learning opportunities are future-forward, adaptable to their goals, needs and circumstances, and equip them to exercise agency in their education and career journeys.

Supporting individualized learning, for example, requires systems that give educators the time, tools and structures to understand and respond to each learner’s growth. Without those conditions, personalization requires extraordinary effort — making it difficult to sustain as a routine part of instructional practice.

Similar structural challenges constrain college- and career-readiness efforts. Educators consistently pointed to the need for more holistic, student-centered pathways. One educator described the importance of a “multi-tiered career program in which students engage in self-exploration of their skills, abilities and interests” to connect learning to concrete opportunities and transferable skills they can use after high school.

Engagement, Agency and the Conditions for Learning

At the crux of learning lies student engagement — shaped by both classroom practices and the broader systems in which learning occurs. Community members and educators both highlighted that academic success depends on students’ well-being.

Students shared that learning is most meaningful when it connects to their interests and allows them to have a voice in shaping their educational experiences. Educators echoed this perspective, underscoring the importance of agency in fostering meaningful learning. As one educator reflected, ensuring educational excellence requires continually redefining educational systems in ways that “give every student access to their own version of success.”

Engagement is not simply a matter of student effort or teacher technique, but a product of the environments and systems that shape learning opportunities.

Learning Does Not Stop at the Schoolhouse Door

Students, families and educators who contributed to Digital Promise’s Challenge Map identified supports that go beyond the schoolhouse, offering insight into the social conditions shaping learning. Suggestions for home stability, physical and emotional safety, and balancing responsibilities inside and outside of school highlight how deeply schooling is intertwined with young people’s lives beyond the classroom.

Other insights were deceptively simple yet profound: One group of students suggested creating regular feedback loops in schools so they could share concerns, inform changes to physical spaces and course offerings, and shape how resources are used. Even these straightforward ideas, however, call for systemic shifts in how schools operate and how student voices are embedded in decision-making.

What It Means to Put People at the Center of Innovation

Education remains a fundamentally human endeavor. As long as the goal is to prepare young people to navigate their futures with skill, agency and well-being, the conditions and relationships that shape students’ opportunity and engagement remain essential.

At a time when education research and development (R&D) is often synonymous with emerging technologies, shifting the focus to problem-solving — driven by the perspectives of those living the challenges — expands what counts as innovation. Existing technologies may play an important role, but they should not be scaled simply because they are novel.

Rather, the starting point for innovation should be: What is the central problem that needs to be solved, for and with whom, and what are the resulting outcomes if the problem is addressed successfully? Only then should existing tools or new solution development enter into the equation. Addressing these challenges requires shifts in mindsets and power dynamics so that both students and educators learn how student voice should shape learning and curriculum.

Why Education Research and Development Needs a Systems Lens

As education R&D evolves, the field is increasingly recognizing that local district systems and community engagement have often been missing from innovation efforts. In policy and education leadership circles, there is a growing call for education R&D that strengthens young people’s futures and, by extension, the nation’s long-term economic and civic well-being.

When schools and local communities are meaningfully engaged in R&D, their perspectives consistently point to persistent challenges that require a systems-level response. These challenges are not isolated problems to be solved with standalone interventions, but signals of deeper misalignments in policies, incentives and assumptions across the education ecosystem.

Questions for Building Lasting Change

Solution developers, policymakers and funders drive change through their respective products and investments. Recognizing these challenges as persistent problems and indicators of necessary systems change, they might consider:

• How well do solutions capture the actual problems they aim to solve, rather than the technological possibilities they allow?
• To what extent do local policies and incentives support the development of solutions that center students, families, communities and educators experiencing the challenge?
• How are the perspectives of those living the challenges incorporated throughout the research, solution design and implementation process?
• How do technological solutions reflect the relational and mindset shifts required across the system?
• How can the evaluation of challenges in education take a systems approach that not only accounts for easily identifiable policies, resources, and practices but also for underlying relationships and assumptions?

Above all, lasting educational innovation depends on a shared conviction: The voices and experiences of students, families, community members and educators must shape how problems are defined and solutions are developed.

Google is rolling out a sizeable update to Gemini for Home, bringing smarter voice controls and a long list of fixes. This update aims at making your smart home feel, well, smarter.

Announced by Google’s Home Chief Product Officer Anish Kattukaran, the update focuses on improving how Gemini understands context — both in what you say and where you say it.

One of the more practical changes is improved device isolation. Now, saying “Turn off the kitchen” will target just the lights. It will not accidentally switch off plugs or unrelated devices. Likewise, “Turn off all the lights” will stay within your home. It will not mistakenly affect other linked locations.

Veteran headphone engineer Axel Grell, best known for developing many of Sennheiser’s most advanced and commercially successful high-end models before launching his own brand, is bringing his latest design to North America. The Grell OAE2 open-back headphone will make its U.S. public debut at CanJam NYC 2026 on March 7–8, marking the model’s first appearance on this side of the Atlantic following its initial German release.

Building on the foundation of the original OAE1, the $599 OAE2 reflects Grell’s continued focus on tonal accuracy, mechanical precision, and long-term listening comfort. The open-back over-ear design incorporates a newly optimized dynamic driver and acoustically refined housing intended to improve airflow and deliver a more speaker-like soundstage presentation in front of the listener.

Grell describes the tuning as natural and neutral, with controlled low frequencies, a lifelike midrange, and extended detail up top without fatigue. Global availability is scheduled for March 31, 2026, priced at $599 / £499 / €499.

Recreating a More Speaker Like Listening Perspective

Designed and engineered in Germany, the OAE2 is built around a clear objective: reducing the “in-head” effect common to many headphones and moving the listening perspective closer to what listeners experience with loudspeakers. Rather than simply creating a wider soundstage, the focus is on reproducing depth, placement, and spatial stability in a way that feels more natural over long listening sessions.

Instead of following the typical open-back headphone layout where the driver fires directly into the ear canal, Axel Grell’s design positions the acoustic output to interact more deliberately with the outer ear. This allows the pinna and surrounding structures to contribute to spatial cues before the sound reaches the ear canal.

The idea mirrors what happens when listening to loudspeakers. Sound reaches the ears only after interacting with the head, outer ear, and upper body, creating subtle timing, phase, and tonal variations that the brain uses to interpret direction and distance. The OAE2 attempts to preserve more of those cues within a headphone format so that the perceived soundstage feels more externalised and stable.

This design philosophy is informed in part by Grell’s ongoing research into spatial hearing and headphone perception in collaboration with Leibniz University Hannover. The research helps guide practical design decisions such as driver positioning, acoustic structure, and overall tuning, with the goal of maintaining coherent imaging, natural treble perception, and controlled low-frequency behaviour.

The intention is not to create exaggerated width or artificial spatial effects. Instead, the OAE2 aims to present music in a way that resembles nearfield loudspeaker listening, where instruments and voices appear positioned in front of the listener rather than inside the head. For listeners accustomed to traditional headphone presentation, the perspective may initially feel different, but the design is intended to become more intuitive as the brain adapts to the spatial cues over time.

Engineering and Build Designed for Long Term Ownership

Beyond its acoustic design, the OAE2 reflects Axel Grell’s view that premium headphones should be durable, serviceable, and built for long-term ownership rather than short product cycles.

At the core of the design is a 40 mm wideband dynamic driver using a bio cellulose diaphragm. The driver works alongside a carefully developed damping system that includes a precision manufactured stainless steel mesh produced in Germany. This combination is intended to maintain controlled airflow and consistent driver behavior while supporting the headphone’s spatial presentation goals.

The physical construction follows a modular approach. The OAE2 uses a fully metal frame and replaceable components that allow the headphone to be maintained and serviced over time. Key parts can be removed or replaced if necessary, extending the usable life of the product and reducing the likelihood that the headphone becomes disposable when individual components wear out.

This design philosophy reflects a broader trend in the high end headphone market. Manufacturers including Meze Audio have helped popularize modular construction, where headphones are designed with replaceable parts and serviceability in mind so they can remain in use for many years rather than being replaced entirely.

Even the packaging follows the same thinking. The OAE2 ships in largely plastic free packaging designed to reduce unnecessary waste while aligning with Grell’s emphasis on sustainability and long term value.

Connectivity, Accessories, and Technical Specifications

The OAE2 is supplied with both balanced and single ended connectivity options, allowing it to work with a wide range of headphone amplifiers, portable players, and desktop audio systems. In the box, listeners will find two detachable cables: a 1.8 metre (5.9 ft) single ended cable terminated with a 3.5 mm plug and a 1.8 metre (5.9 ft) balanced cable with a 4.4 mm connector. A screw on 3.5 mm to 6.3 mm adapter is also included for compatibility with traditional headphone amplifier outputs, along with a protective carry case for storage and transport.

Technically, the OAE2 is a circumaural open-back headphone built around a dynamic transducer. Although specifications are nearly identical to the OAE1, the OAE2 are 3 grams heavier with an even wider frequency response rated from 12 Hz to 34 kHz within a ±3 dB window, extending from 6 Hz to 46 kHz at -10 dB. Nominal impedance is the same at 38 ohms with a sensitivity of 100 dB at 1 kHz (1 VRMS), suggesting that the headphone can be driven by a variety of modern sources while still benefiting from a capable headphone amplifier. Total harmonic distortion is rated at 0.05 percent at 1 kHz and 100 dB. The headphone weighs 378 grams (13.3 oz) without the cable attached.

The Bottom Line

The OAE2 represents the next step in Axel Grell’s effort to rethink how headphones present space. Rather than chasing exaggerated width or DSP tricks, the design focuses on repositioning the driver and shaping the acoustics so the sound appears more in front of the listener, and closer to how nearfield speakers behave. It’s a concept Grell has been refining for several years, and we first heard an early prototype of this approach at CanJam SoCal in 2023, where the forward projecting presentation immediately stood out from the usual “inside your head” headphone experience.

The new model builds on the earlier OAE1 concept but arrives with a more mature design and a higher price point at $599, compared to the roughly $300 launch price of the original. That shift reflects a more fully developed product with revised acoustics, upgraded construction, and a clearer articulation of Grell’s spatial listening philosophy.

Listeners who prioritize tonal balance, realistic imaging, and long term listening comfort are the most likely audience here. Those expecting the traditional wide but internalized headphone stage may find the presentation different at first, but the goal is a more natural spatial perspective that resembles listening to speakers at close range.

We’ll have the chance to spend more time with the OAE2 during its North American debut at CanJam NYC 2026, where Grell will be demonstrating the headphone publicly for the first time in the U.S. If all goes according to plan, we expect to have a full review ready before the end of March once production units become available.

For more information: grellaudio.com

Related Reading:

Microsoft has released the KB5075039 Windows Recovery Environment update for Windows 10 to fix a long-standing issue that prevented some users from accessing the Recovery environment.

The Windows Recovery Environment (WinRE) is a minimal troubleshooting environment used to repair or restore the operating system after it fails to start, to diagnose crashes, or to remove malware.

In October 2025, Microsoft confirmed that the KB5066835 Patch Tuesday updates broke USB mouse and keyboard input when using the Windows 11 Recovery Environment, making it difficult for many to use the troubleshooting tool.

While they quickly rolled out a fix for this flaw, they didn’t disclose until February that the Windows 10 KB5068164 update released in October also broke WinRE.

“This update contains an issue that prevents the Windows Recovery Environment from starting successfully,” reads the February update to the change log.

Yesterday, Microsoft released the “KB5075039: Windows Recovery Environment update for Windows 10” to fix the WinRE issue introduced last year.

“[Windows Recovery Environment (WinRE)] Fixed: WinRE would not start after installing the October 14, 2025 update KB5068164,” reads the change log.

To install the update, your WinRE partition must be at least 256MB in size. If not, you will need to increase the partition size using these instructions.

Before resizing any partition, including the WinRE partition, it is always advisable to back up the data on the drive whose partitions are being resized.

Malware is getting smarter. The Red Report 2026 reveals how new threats use math to detect sandboxes and hide in plain sight.

Download our analysis of 1.1 million malicious samples to uncover the top 10 techniques and see if your security stack is blinded.

Download The Report

A new device aims to give people control over who can hear them in a world filled with gadgets that are always listening and capturing your conversations. A startup called Deveillance has introduced Spectre I, a portable device designed to stop microphones in nearby devices from recording your voice.

The company says the device can make conversations unintelligible to phones, smart speakers, laptops, and other gadgets that constantly listen for audio. The idea addresses a growing concern around always-on devices.

According to the company, about 14.4 billion devices worldwide are continuously listening for voice input. These recordings often become valuable data sources used for data mining, training artificial intelligence systems, influencing our buying behaviours or deepest opinions.

Even a short sample of speech can reveal sensitive personal details. Around 30 seconds of voice data can help determine traits such as age, weight, income level, and even health information.

A device that creates a privacy bubble around your voice

Spectre I works by creating a two meter protection zone around the user. When activated, it scans for nearby microphones and emits signals that humans cannot hear, but microphones can detect.

These signals overlay your speech so that recording devices receive distorted audio that cannot be understood.

Unlike traditional signal jammers that rely on strong radio interference, the device uses artificial intelligence, signal processing, and physics based research to target microphones directly.

The system operates locally on the device and does not send any data to the cloud. The portable design of Spectre I makes it easy to carry anywhere.

Deveillance says this makes it useful in business meetings, personal conversations, or any situation where people want to keep discussions private.

The company has opened pre-orders for Spectre I with a refundable deposit of $1,199. The device is currently in development, with the first shipments expected in the second half of 2026.

Privacy groups like the Electronic Frontier Foundation have long warned about the risks of always-on surveillance. Deveillance says Spectre I is only the beginning of its effort to give users more control over how their data is collected and shared.

It’s been rumoured for a long time, but Apple has finally taken the wraps off arguably its most exciting laptop in years.

The $599 MacBook Neo arrives as the most affordable entry in Apple’s laptop range, with a price that’s more in line with the brand’s iPad range – including the new iPad Air M4 – than the MacBook Air or MacBook Pro.

This is the first MacBook to be powered by the A18 Pro chip originally made for an iPhone, and, like the iMac, it also comes in a range of fun colours – Silver, Indigo, Blush, and Citrus – that are also a first for a MacBook. Apple has colour-matched the keyboard and feet, giving it a very distinct look.

The A18 Pro chip has a 6-core CPU (with 2 performance cores and 4 efficiency cores) and a 5-core GPU, along with ray-tracing support and a 16-core neural engine. Apple will only offer a single 8GB memory option, with storage sizes of either 256GB or 512GB.

Advertisement

Advertisement

Battery life is stated at 16 hours for video streaming and 11 hours of web browsing, and there’s a 1080p camera on the front. Unlike the other MacBook models, there’s no notch, but a bezel similar to that of an iPad. The display is 13 inches, with a 2408 x 1506 resolution and a reported 500 nits of brightness.

The base model ships without Touch ID, although you can pay a little more and get a version with the fingerprint unlock embedded into the keyboard.

Connectivity comes from two USB-C (one is a USB 2 and the other USB 3) ports and a headphone jack, so no MagSafe charging. There is Wi-Fi 6E and Bluetooth 6 though, which is welcome.

MacBook Air Neo Price and Release Date

Prices start at £599/$599 for a model with 256GB storage and £699/$699 for a Touch ID-toting 512GB variant. It can only be selected with 8GB memory, and there’s no 1TB storage option.

Advertisement

This is a breaking news story. We’ll update it as we get more information

Advertisement

To create coherent images or videos, generative AI diffusion models like Stable Diffusion or FLUX have typically relied on external “teachers”—frozen encoders like CLIP or DINOv2—to provide the semantic understanding they couldn’t learn on their own.

But this reliance has come at a cost: a “bottleneck” where scaling up the model no longer yields better results because the external teacher has hit its limit.

Today, German AI startup Black Forest Labs (maker of the FLUX series of AI image models) has announced a potential end to this era of academic borrowing with the release of Self-Flow, a self-supervised flow matching framework that allows models to learn representation and generation simultaneously.

By integrating a novel Dual-Timestep Scheduling mechanism, Black Forest Labs has demonstrated that a single model can achieve state-of-the-art results across images, video, and audio without any external supervision.

The technology: breaking the “semantic gap”

The fundamental problem with traditional generative training is that it’s a “denoising” task. The model is shown noise and asked to find an image; it has very little incentive to understand what the image is, only what it looks like.

To fix this, researchers have previously “aligned” generative features with external discriminative models. However, Black Forest Labs argues this is fundamentally flawed: these external models often operate on misaligned objectives and fail to generalize across different modalities like audio or robotics.

The Labs’ new technique, Self-Flow, introduces an “information asymmetry” to solve this. Using a technique called Dual-Timestep Scheduling, the system applies different levels of noise to different parts of the input. The student receives a heavily corrupted version of the data, while the teacher—an Exponential Moving Average (EMA) version of the model itself—sees a “cleaner” version of the same data.

The student is then tasked not just with generating the final output, but with predicting what its “cleaner” self is seeing—a process of self-distillation where the teacher is at layer 20 and the student is at layer 8. This “Dual-Pass” approach forces the model to develop a deep, internal semantic understanding, effectively teaching itself how to see while it learns how to create.

Product implications: faster, sharper, and multi-modal

The practical results of this shift are stark. According to the research paper, Self-Flow converges approximately 2.8x faster than the REpresentation Alignment (REPA) method, the current industry standard for feature alignment. Perhaps more importantly, it doesn’t plateau; as compute and parameters increase, Self-Flow continues to improve while older methods show diminishing returns.

The leap in training efficiency is best understood through the lens of raw computational steps: while standard “vanilla” training traditionally requires 7 million steps to reach a baseline performance level, REPA shortened that journey to just 400,000 steps, representing a 17.5x speedup.

Black Forest Labs’ Self-Flow framework pushes this frontier even further, operating 2.8x faster than REPA to hit the same performance milestone in roughly 143,000 steps.

Taken together, this evolution represents a nearly 50x reduction in the total number of training steps required to achieve high-quality results, effectively collapsing what was once a massive resource requirement into a significantly more accessible and streamlined process.

Black Forest Labs showcased these gains through a 4B parameter multi-modal model. Trained on a massive dataset of 200M images, 6M videos, and 2M audio-video pairs, the model demonstrated significant leaps in three key areas:

1. Typography and text rendering: One of the most persistent “tells” of AI images has been garbled text. Self-Flow significantly outperforms vanilla flow matching in rendering complex, legible signs and labels, such as a neon sign correctly spelling “FLUX is multimodal”.

2. Temporal consistency: In video generation, Self-Flow eliminates many of the “hallucinated” artifacts common in current models, such as limbs that spontaneously disappear during motion.

3. Joint video-audio synthesis: Because the model learns representations natively, it can generate synchronized video and audio from a single prompt, a task where external “borrowed” representations often fail because an image-encoder doesn’t understand sound.

In terms of quantitative metrics, Self-Flow achieved superior results over competitive baselines. On Image FID, the model scored 3.61 compared to REPA’s 3.92. For video (FVD), it reached 47.81 compared to REPA’s 49.59, and in audio (FAD), it scored 145.65 against the vanilla baseline’s 148.87.

From pixels to planning: the path to world models

The announcement concludes with a look toward world models—AI that doesn’t just generate pretty pictures but understands the underlying physics and logic of a scene for planning and robotics.

By fine-tuning a 675M parameter version of Self-Flow on the RT-1 robotics dataset, researchers achieved significantly higher success rates in complex, multi-step tasks in the SIMPLER simulator. While standard flow matching struggled with complex “Open and Place” tasks, often failing entirely, the Self-Flow model maintained a steady success rate, suggesting that its internal representations are robust enough for real-world visual reasoning.

Implementation and engineering details

For researchers looking to verify these claims, Black Forest Labs has released an inference suite on GitHub specifically for ImageNet 256×256 generation. The project, primarily written in Python, provides the SelfFlowPerTokenDiT model architecture based on SiT-XL/2.

Engineers can utilize the provided sample.py script to generate 50,000 images for standard FID evaluation. The repository highlights that a key architectural modification in this implementation is per-token timestep conditioning, which allows each token in a sequence to be conditioned on its specific noising timestep. During training, the model utilized BFloat16 mixed precision and the AdamW optimizer with gradient clipping to maintain stability.

Licensing and availability

Black Forest Labs has made the research paper and official inference code available via GitHub and their research portal. While this is currently a research preview, the company’s track record with the FLUX model family suggests these innovations will likely find their way into their commercial API and open-weights offerings in the near future.

For developers, the move away from external encoders is a massive win for efficiency. It eliminates the need to manage separate, heavy models like DINOv2 during training, simplifying the stack and allowing for more specialized, domain-specific training that isn’t beholden to someone else’s “frozen” understanding of the world.

Takeaways for enterprise technical decision-makers and adopters

For enterprises, the arrival of Self-Flow represents a significant shift in the cost-benefit analysis of developing proprietary AI.

While the most immediate beneficiaries are organizations training large-scale models from scratch, the research demonstrates that the technology is equally potent for high-resolution fine-tuning. Because the method converges nearly three times faster than current standards, companies can achieve state-of-the-art results with a fraction of the traditional compute budget.

This efficiency makes it viable for enterprises to move beyond generic off-the-shelf solutions and develop specialized models that are deeply aligned with their specific data domains, whether that involves niche medical imaging or proprietary industrial sensor data.

The practical applications for this technology extend into high-stakes industrial sectors, most notably robotics and autonomous systems. By leveraging the framework’s ability to learn “world models,” enterprises in manufacturing and logistics can develop vision-language-action (VLA) models that possess a superior understanding of physical space and sequential reasoning.

In simulation tests, Self-Flow allowed robotic controllers to successfully execute complex, multi-object tasks—such as opening a drawer to place an item inside—where traditional generative models failed. This suggests that the technology is a foundational tool for any enterprise seeking to bridge the gap between digital content generation and real-world physical automation.

Beyond performance gains, Self-Flow offers enterprises a strategic advantage by simplifying the underlying AI infrastructure. Most current generative systems are “Frankenstein” models that require complex, external semantic encoders often owned and licensed by third parties.

By unifying representation and generation into a single architecture, Self-Flow allows enterprises to eliminate these external dependencies, reducing technical debt and removing the “bottlenecks” associated with scaling third-party teachers. This self-contained nature ensures that as an enterprise scales its compute and data, the model’s performance scales predictably in lockstep, providing a clearer ROI for long-term AI investments.