News

• Microsoft is eating humble pie over its unrealistic RAM recommendations
• It has deleted articles that pushed 32GB as an ideal or 'no-worries' loadout
• Apple is also feeling the heat in the RAM crisis, with rumors that it's struggling to secure an alternative source of memory supply from China

There are some fresh twists with the RAM crisis hitting some big tech companies, as Microsoft has backtracked on its previous memory recommendations, and even Apple is apparently finding it difficult to cope with the scarcity of memory.

Let's discuss Microsoft first, and as Windows Latest pointed out, the company has been busy backpedalling on previous memory recommendations now that the RAM crisis – which just keeps getting worse – has made those suggestions look foolish.

Microsoft previously had support documents in its Windows Learning Center which have now been removed, and Windows Latest highlights two of them. One was about optimizing your gaming PC, and it advised that "32GB is ideal for serious players who run the most demanding titles" (albeit the article also said 16GB was "plenty" for most games).

Another piece said that 32GB of RAM was the "no-worries zone", and that article was also quietly deleted as there was some backlash against this, given that it was published when the price of system memory had become ridiculous. (And buying a 32GB kit was very much a worry for your wallet).

The links to those articles now redirect to the home page of the Learning Center, and Microsoft is evidently trying to forget about pushing 32GB of RAM as an 'ideal' or 'worry-free' target for memory on your PC.

More broadly, since Copilot+ PCs were launched and the AI features for these devices made them require 16GB, Microsoft has obviously been keen to have that as a baseline memory configuration. Except now, a change of stance is necessary, as with the RAM crisis reaching alarming new heights, Microsoft has been forced to enact huge price hikes with its Surface devices.

And of course, the latest twist with that Surface hardware is that Microsoft has brought back 8GB models with last year's Surface Pro and Surface Laptop. Which makes it kind of difficult to push 16GB as a minimum, let alone make suggestions that 32GB is where it's really at for properly smooth performance.

The abandonment of these Learning Center articles is hardly surprising, then, and Microsoft is also addressing how speedily Windows 11 runs with 8GB of memory (not quickly enough currently). One of its promises with fixing the OS was better performance with a leaner RAM loadout, and Microsoft just made it clear that the company is now actively working to make Windows 11 run better with 8GB before the end of the year.

This has become a vital goal, really, when you consider that Apple has pulled off a commendable showing of performance with its MacBook Neo that packs 8GB of RAM. That was effectively a gauntlet thrown down for Microsoft – something of a declaration that macOS is coming to try to take Windows 11's market share – and one that the Windows maker had to respond to (which became clear enough when Microsoft went on the attack against the Neo).

Speaking of Apple, Microsoft isn't the only tech giant being buffeted by the rising costs caused by the RAM storm. While it has had a big success with the Neo, the challenge for Tim Cook's firm – soon to be John Ternus's, of course – is to maintain that momentum, and by all accounts, that's proving a tricky task.

Recent Mac price hikes have caused a good deal of pain – taking some of the wind out of the good ship Neo's sails – and Apple is trying to secure its RAM supply lines for the future, with rumors abounding that it's turning to Chinese chip makers to find extra production capacity.

The latest speculation, however, is that according to a report from Digital Daily (a Korean tech site, via Wccftech), Apple has floundered in negotiations with Chinese memory giant CXMT.

Apparently CXMT has strong enough domestic demand that it doesn't have to offer more attractive pricing to Apple. Essentially, CXMT is holding the line and has "insisted on prices that were higher or similar to those offered by Samsung or SK Hynix" (bear in mind there may be nuances lost with the translation of the article).

You get the message, though: Apple is failing to obtain better deals on mobile DRAM, which includes LPDDR5X, for its iPhones (and that RAM is also used in its MacBooks, of course). And CXMT was supposed to be an escape route from the difficulties of getting enough RAM inventory from Samsung and SK Hynix (and also Micron, a key supplier for Apple), but it seems like a dead-end for now.

At least if this report is correct, but analyst Tim Culpan has also written a short post which backs up the notion that Apple is in trouble here. Culpan writes: "Apple and its suppliers are scrambling to get enough memory chips for its upcoming release of new iPhone models."

'Scrambling' is a word that evokes quite a sense of panic, and indeed with the iPhone 18 models – and the foldable offering – not much more than a month away from launch now (in theory), I'd bet there are some heated words flying here and there. Culpan notes: "Assemblers are working with Apple to rush shipments of DRAM used in mobile devices."

That report is specifically about smartphones, mind, but this same situation applies to mobile RAM that's also used in MacBooks.

The overall theme is more RAM misery all round, which is hardly a surprise given all the negative news we've been hearing on the grapevine of late. GPU pricing has been the latest round of doom and gloom over the past week or two, and I don't think the pessimistic news is going to stop flowing for the foreseeable.

It's not bad news for everyone, though. A source in the semiconductor industry told Digital Daily that: "With the general-purpose DRAM floor remaining unbroken and Samsung and SK Hynix monopolizing the lead in high-value AI memory such as HBM4, the operating profit margins and global market control of the domestic semiconductor sector are expected to rise even more steeply in the second half of the year."

So, it's a familiar story: profits will be on the up and up for memory makers, while consumers will be suffering the pain of the hikes. Apple's purported rush for RAM supplies as the clock runs down on the iPhone launch window is a particularly worrying sounding story, one that doesn't bode well in terms of avoiding more Mac (and iPhone) price rises in the future. Neither can we rule out more Surface price rises, or other laptops for that matter.

The enterprise race to scale AI operations is in full swing – both from a deployment and security perspective. While conventional wisdom suggests that the teams who deploy the models first get the edge, it’s the wrong approach for enterprises.

For nefarious actors, speed is the name of the game. Malicious actors leveraging AI to probe for vulnerabilities don’t need to share their decision trail to audit committees or regulators – making speed alone the key advantage for attackers.

Enterprise security teams, on the other hand, operate under entirely different parameters – which also happen to be where the opportunity lies.

Security teams don’t just need the capability to identify anomalies, screen transactions, or make access decisions – they also need to be prepared to explain what happened, when, and why to key stakeholders.

Every action and every outcome needs to be clarified and justified to a board, an auditor, or a customer.

Scaling AI at the enterprise level is not relegated to who moves fastest, but rather who can embed the necessary accountability frameworks, emergency brakes, and audit trails.

In truth, speed is not the primary challenge for most enterprise teams. The bigger, more difficult challenge is found in data and governance. Policy models and detection frameworks are only as impactful as the information inputs. However, most organizations are devoid of structured, well-governed data – particularly as it pertains to AI activity. Across the majority of organizations, data is scattered.

Activity logs, access records, transaction histories all live in disparate systems with inconsistent formats and no single source of truth. No matter how much of this data is fed into an AI model, clarity will never be achieved. Instead, teams generate a false sense of confidence built upon a shaky foundation.

Meaningful scale begins with auditability and accountability as infrastructure, not an afterthought. As AI systems become increasingly embedded across organizations, the requirement heightens for more consistent records of actions that occurred and policy enforcement where actions are checked against defined rules prior to execution.

Equally as important, organizations need to have the internal muscle memory to explain automated decisions to stakeholders outside the engineering team, because they've been doing versions of this for years in compliance and risk functions.

When this approach is foundational for enterprises, AI systems become genuinely useful. AI models built upon strong data and governance standards can meaningfully screen actions against known risks before settlement, flag patterns that humans may have missed, and maintain updated records that can be shared across key stakeholder groups. Devoid of this foundation, fragmented data and ungoverned systems compound – resulting in faster mistakes.

Enthusiasm around AI systems centers on the deployment side of the equation, without asking whether the underlying infrastructure can support what's being layered on top of it. This approach is suboptimal and leaves an open door for breaches – akin to installing an alarm system in a building without putting locks on the doors.

Embedding infrastructural policy enforcement, audit trails, and human-reviewable records does bring an additional layer prior to deployment. And in a domain where nefarious actors operate free of this operational layer, it's fair to ask whether enterprise security teams are creating a permanent speed disadvantage for themselves.

However, that mindset misses what the added operational layer actually delivers: the difference between an AI system that fails safely and one that fails silently.

Enterprise security teams who operate slower but can consistently and proactively identify and reverse bad decisions are in a fundamentally different position than enterprises who operate quickly and discover a failure three weeks post-incident. The tradeoff is real, but it’s the wrong tradeoff to optimize away.

A common misconception is that security decisions are made for and by security teams. That’s not the case. Security decisions made by automated systems need to satisfy stakeholders far outside of this workstream – including regulators, insurers, customers, even boards.

Those stakeholders aren’t moved by sophistication. They care whether the organization can clearly and consistently demonstrate what the system did and why. Organizations without that operational layer find that AI adoption increases their potential risk exposure, because they're now making faster decisions with insufficient guardrails.

As AI systems continue to increasingly exhibit autonomous behaviors, the stakes continue to change. An AI system that can initiate payments, approve transactions, or move funds independently doesn't just need a policy to follow – it needs brakes to press in the event of a bad decision.

Without this layer embedded, enterprise security teams not only carry the accountability problem – they also have no chance of catching a mistake before it becomes permanent.

Readiness requires the proper sequencing. Before scaling AI operations, it’s critical to make an honest assessment of three things: whether underlying data across systems are structured, whether there are firm policy checks in place, and whether verifiable records of automated decisions can be produced on-demand.

Starting here positions AI to become a force multiplier for security teams. These three elements make all the difference between catching what humans miss and doing it fast enough to matter – or simply adding velocity to a faulty process.

The “AI race” will not be won by having the newest models. Really, it’s about devoting effort to the unglamorous work – building the data infrastructure and governance systems that make AI models trustworthy.

We've reviewed, rated, and ranked the best firewall software.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

• Google’s automated systems mistakenly flagged hundreds of Blogger sites as malicious
• Company admitted a bug caused false malware labels, promising a fix
• Users advised to request reviews, avoid migrating content

Hundreds of Blogger websites were locked down, and some apparently deleted as well, after Google’s automated security systems erroneously flagged them as malicious.

A user posted a new message on Google’s forum saying the huge number of reports regarding locked blogs are all for the same reason - Malware and Similar Malicious Content.

The nature of the lockdown “suggests misclassification by automated systems”, the post reads, adding that the team has “already been notified of this issue.”

Those affected will see a red padlock in their dashboard and a warning saying the blog was locked:

"This blog was removed for violating Blogger's Community Guidelines. If you wish to request a review of the blog, click 'Request Review' below," the notice reads.

At press time, the forum post had more than 500 “I have the same question” votes, and more than 200 replies.

In a statement given to BleepingComputer, Google said it was aware of a bug that falsely labeled many sites as malicious, and that it was working on a fix.

"We are aware of a bug that incorrectly flagged some Blogger-hosted sites as malware for less than a day. We are working on a fix to resolve the issue as quickly as possible," the company said.

To make matters even worse, Google said that if users don’t file an appeal, that their blogs can be permanently deleted.

Users are advised not to create new blogs and migrate content, since that is in violation with Google’s TOS. They are also advised against deleting their Blogger profile or service from their Google account, since this will irrevocably delete the blogs. They can, however, back up their blogs if they are afraid of losing the content.

The full extent of the issue is unknown, but according to BleepingComputer, the number of users on the platform exceeds 200,000.

A reckoning is underway in enterprise AI. Chief executives who spent two years bolting frontier models onto their businesses are now asking harder questions.

What did we actually get for the money? Where does our data go when it flows through someone else's model? And when the model we standardized on last year isn’t the best one this year, how much of our business have we quietly handed over to a vendor we don't control?

That last question is the one that should keep leaders up at night, because for most companies, the honest answer is: far more than they think.

The conversation about AI has mostly been about which model is best.

That’s the wrong question.

In a field where leadership changes hands every few quarters, “best model” is a snapshot, not a strategy.

The question that actually matters is architectural: when the state of the art moves — and it always does — can you move with it, or do you have to rebuild your business every time?

Here's what a decade of running AI at production scale teaches you that no benchmark chart will: models are disposable, and they're getting more disposable by the year. The transcription engine that led the market when we started is a footnote now.

The computer vision system that dominated three years ago has been lapped over and over. We've swapped best-in-class engines in and out of live customer workflows hundreds of times — across speech recognition, translation, object detection, face redaction, and now large language models — and each cycle turns over faster than the one before it.

An company that’s hard-wired to one provider inherits that provider's roadmap, pricing, and priorities as its own. When the provider raises prices, you pay. When it deprecates the version you built on, you rebuild.

When a smaller open-source model, fine-tuned on your own data, would actually do the job better and cheaper, you can't reach for it because your workflows only speak one dialect. That isn't a partnership. It's a dependency — and depending on a moving target is about the most expensive position you can be in.

That's the reasoning behind building systems where model lock-in isn't an option from day one: an orchestration layer that connects and manages hundreds of commercial, open-source, and proprietary models across different cognitive tasks, routes each job to whatever engine is best for it, and swaps models out as the state of the art shifts — without anyone having to rebuild a workflow.

The model becomes a component, not a foundation. The real foundation is the orchestration layer and the data underneath it, and those stay owned by the enterprise, not the vendor.

Model plurality sounds good in a keynote and collapses in practice without one thing: the ability to prove, on your own data, which model is actually better for your job. “Best” isn’t a leaderboard position.

A model that tops a public benchmark can badly underperform on your accents, your camera angles, your legal thresholds, your definition of “good enough.” Public benchmarks measure general capability.

They tell you almost nothing about how a model will perform inside your specific workflow. So the real currency of the next era isn't the model — it's the evaluation.

The companies pulling ahead are the ones who can put any engine up against any other on their own content, with their own quality bar, and make swap decisions based on evidence instead of vendor marketing.

That means scoring engines against each other continuously, on real customer data, so “which model” stays a measured decision you can remake any time the field shifts — not a one-time choice you're stuck with.

That evaluation muscle is itself a strategic asset. It's what turns a pile of interchangeable models into a compounding advantage — and it's precisely the capability an enterprise forfeits the moment it standardizes on a single black box.

There's a reason this matters most in audio and video. Text is basically commoditized at this point. The proprietary, defensible, hard-to-replicate data in the enterprise is the recorded record of what a company actually said, did, made, and witnessed — decades of broadcast, footage, calls, and captured events. It's multimodal, it's rights-encumbered, and it can't be replaced, which also happens to make it exactly what this generation of AI is hungriest for.

The appetite for training and tuning data has outrun what the open web can supply. What's actually needed now is what enterprises already have sitting in their archives: vast, rights-cleared, real-world, multimodal data, plus expert human judgment about what “good” looks like.

Which makes the default posture of the last two years perverse — companies have paid premium prices to push their proprietary audio and video through third-party models, with limited visibility into what's retained, learned, or one day competed against them. If your data is the scarce input everyone's after, the last thing you want to do is hand it over as a byproduct of your software bill.

The alternative is building the enterprise business the other way around: turning an organization's raw archives into AI-ready, enriched assets it actually owns, with rights and governance metadata baked in at the point of creation rather than tacked on afterward — because when the data in question is a witness's voice, an athlete's likeness, or a rights-encumbered broadcast, provenance and consent aren't optional extras.

From there, rightsholders can put that data to work themselves — licensing it to model developers and cloud providers on their own terms, with consent, provenance, and compensation built into the deal.

Watch where the sophisticated buyers are going and the pattern is unmistakable The old reflex — route everything to whichever frontier model is biggest — is giving way to something more deliberate: a portfolio approach where smaller, open-source, and fine-tuned models handle most of the day-to-day workloads, and the giant models get reserved for the problems that genuinely need them.

The reasons are practical — cost, latency, data control, and the ability to specialize a model on proprietary data until it beats a general-purpose giant at your specific task.

This is where the two ideas come together. Fine-tuning and open source only work in your favor if you actually own the data to tune on and have the architecture to deploy into. A company locked to one vendor can't fine-tune an open model on its own footage and slot it into production as the plumbing simply won't allow it.

An enterprise with an orchestration layer and governed, AI-ready data can do exactly that, and can keep doing it as better base models emerge. Owning your data and having freedom in your architecture are what actually make this new era of specialized, fine-tuned, open models available to you at all. Without them, you're watching a shift from the sidelines that was supposed to be your advantage.

It’s okay to be wary of how fast “sovereign AI” is becoming a marketing category, because sovereignty delivered through a new single-vendor dependency is just lock-in with better branding. Real sovereignty is an architectural posture with three commitments.

First, model plurality, proven by evaluation: every model — commercial, open-source, or fine-tuned — tested on your data and replaceable at will.

Second, governance that travels with the data: provenance, auditability, consent, and policy enforced at the data layer.

Third, actual ownership: your audio and video, enriched and controlled as an asset you deploy on purpose, not one that leaks out incidentally. None of this is an argument against the frontier labs — they build extraordinary technology, and plenty of companies use it every day, through architecture that keeps the leverage on their side of the table.

The enterprise doesn't have to choose between using the best models in the world and controlling its own future. The whole point is to do both: orchestrate every model worth using, prove which one wins on your own data, and own the audio, video, and judgment that make any of them worth running.

The model is temporary. Your data and your judgment are not. Build for that, and you spend the next decade compounding an advantage your competitors rented and lost.

We tested out the best data recovery software for Mac and PC.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit