• Micron CEO says company is unable to meet current demand
• DRAM production is being prioritized for AI and datacenters
• Consumers are reeling for the cheap RAM of yesteryear

Micron Technology Inc. CEO Sanjay Mehrotra has said that the company “are only able to supply, for our key customers in the midterm, about 50% to two-thirds of their requirements.”

Mehrotra’s statement reflects the growing demand by datacenters for components related to AI compute that will likely worsen the supply of memory.

from the the-kids-are-alright dept

We’ve been covering Australia’s under-16 social media ban since before it went into effect, first noting the confusion and obvious implementation problems as pretty much everyone realized it was a total mess, and then documenting how the ban was actively harming kids with disabilities by cutting them off from critical support communities.

None of this was even remotely surprising. Critics around the world warned about all of it. The government went ahead anyway because doing something tends to poll better than doing something that actually works, especially when the thing that works is harder to explain. And government officials insisted (incorrectly) that the only ones who were complaining were the big tech companies or their proxies.

Now, three months in, the data is starting to arrive, and it confirms what should have been obvious from the start. New data from parental monitoring company Qustodio, provided to Crikey, shows that the ban has barely moved the needle:

While TikTok, YouTube and Snapchat all saw a decrease in use by Australians aged 10-15, the majority of teens who had been using the social media platforms pre-ban remained on the services afterwards.

That’s according to a new snapshot of data provided to Crikey by parental monitoring company Qustodio, adding to early evidence that there’s widespread circumvention of the government’s flagship tech policy.

Advertisement

The usage drop was only marginally larger than the normal seasonal dip that happens every year. In other words, the “world-first” ban achieved roughly the same effect as summer ending. There was definitely a drop, but it’s just not a particularly big one:

For what it’s worth, others are reporting the same thing. The Courier Mail found that the majority of teens who were using these apps before the ban were still using them afterwards.

Defenders of the ban will usually say something along the lines of: “We had to do something. Children were at risk. Even if it’s imperfect, at least we tried.” That argument might hold some water if the ban merely failed — if it just didn’t work and left things roughly where they were before. A swing and a miss. You dust yourself off and try something else.

But that’s not what happened. The ban didn’t leave things where they were. It made things actively worse, through a mechanism that was entirely predictable.

The ban is basically a test of technical sophistication, rather than a test of vulnerability. The kids who can’t figure out how to get around it — or who don’t have friends or older siblings to help them — are the kids who are already isolated or lack the technical skills to bypass a block. Those are the kids with disabilities who lost their support communities, the ones we wrote about last month. Those are the kids in rural areas or difficult home situations who relied on these platforms for connection. The ban selected for vulnerability and filtered against resourcefulness.

That’s a hell of a result for a child safety measure.

Meanwhile, the vast majority of kids — the ones the ban was supposedly protecting — just learned to route around it. Rather than learning responsible usage and digital literacy, they learned that age verification systems are obstacles to be defeated… which, congratulations, is probably the single least useful lesson you could teach a teenager about their relationship with technology.

Actually, it’s worse: Australian adults now have a false sense of security — the comfortable belief that they’ve magically protected kids from the evils of the internet.

When you pass a ban and declare the problem solved, you eliminate the political pressure to do the things that would actually help. Why fund digital literacy programs when kids aren’t supposed to be on social media at all? Why push platforms to develop better age-appropriate tools and experiences when under-16s are “banned”? Why have conversations with kids about healthy usage of something they’re not supposed to be using?

The ban creates a fiction — kids are off social media — that every politician and regulator has an incentive to maintain, even though the data says the fiction is exactly that. Kids are still using these platforms. They’re just doing it without guidance or access to real safety tools, and with the realization that the adults in charge don’t actually understand how any of this works.

So you end up with the worst possible outcome: nearly universal continued usage combined with policy complacency and zero institutional incentive to teach kids how to use these platforms safely. Kids using social media without supervision or education, while the government pats itself on the back for a ban that exists only on paper.

This was all foreseeable. It was all foreseen. Critics said so publicly, repeatedly, before the law passed. And the Australian government did it anyway, because “ban the thing” is a satisfying political narrative, even when — especially when — it doesn’t work.

And now that it’s failed, rather than admit that the plan was bad and dangerous… they’re doubling down by blaming the tech companies:

An eSafety spokesperson said that social media platforms need to take “continuous action” to find underage users on their platforms, including those who’ve created new accounts.

“eSafety is aware of reports some under-16s continue to access social media accounts and is actively engaging with platforms and their age assurance providers to probe weaknesses and encourage continuous improvement of implementation and settings while continuing to monitor for any systemic failures that may amount to a breach of the law,” they said.

The spokesperson foreshadowed further announcements in the coming weeks, adding: “We will provide further updates on age restricted platforms’ progress in meeting their obligations when it is appropriate to do so but we must be careful to not compromise the regulatory process currently underway or prejudice any enforcement action we may undertake in future.”

The blame will keep flowing toward the platforms. The kids will keep routing around the ban. And the adults will keep congratulating themselves for solving a problem they made worse.

Filed Under: australia, esafety, esafety commissioner, kids, safety, social media, social media ban, teens

• Agentic AI enables PCs to autonomously execute multiple tasks in parallel
• Persistent local AI reduces reliance on cloud computing for sensitive workflows
• Professionals can delegate urgent tasks and wake to completed project summaries

The personal computer has been central to work and creativity for four decades, allowing users to write, build, design, and analyze with professional-grade tools – but PCs have largely remained tools operated directly by humans, opening apps and performing tasks manually.

However, AMD now suggests that Agentic AI could be the killer app for even decades-old PCs, transforming them into systems that autonomously execute tasks and manage workflows.

Google is working on a feature in Android 17 that could quietly remove one of the most annoying security steps on your phone. If you use a SIM PIN, you may soon not have to remember it or enter it every time you restart your device.

According to Android Authority, the feature, called automatic SIM lock protection, first appeared in Android 17 beta and is now live in the latest Canary build.

How does automatic SIM lock protection work?

A SIM PIN is different from your phone’s unlock PIN. It protects your SIM card itself and is required when you reboot your phone or insert the SIM into another device. Without it, your SIM cannot be used for calls, texts, or mobile data.

To set up the new feature, you enable Automatic PIN management, confirm your identity with your passcode or biometrics, and then enter your SIM’s current PIN. If you have not set one, you can use your carrier’s default code, which is usually something simple like 0000, 1111, or 1234.

You can find it by heading to Security and Privacy, > More security & privacy > Protect SIM card. You can also view the stored PIN inside the settings by using the ‘Show Android managed PIN’ option.

Once this is done, Android takes over. When you restart your phone and unlock it, the system automatically enters the SIM PIN for you.

Why does this matter?

The biggest issue with SIM PINs has always been convenience. You are already juggling multiple passwords, so remembering another code feels like a hassle. Automatic SIM lock protection removes that burden while keeping the protection in place.

Importantly, the SIM PIN still works as intended if your card is moved to another phone. In that case, the PIN must be entered manually, which helps protect your number from misuse in case of theft.

That makes this feature especially useful if you are worried about someone accessing your calls, messages, or two-factor authentication codes. It is a small change, but one that could make SIM level security more practical for everyday use.

OpenAI is working toward creating a desktop “superapp” that will consist of its three tools: ChatGPT, the coding platform Codex and the Atlas browser, according to a report by The Wall Street Journal. OpenAI executives said the goal behind this new desktop app is to improve the user experience.

The move comes after the Journal reported earlier this week that OpenAI CEO of applications Fidji Simo told employees the company wanted to focus on its core business instead of side projects.

In a Thursday memo to staff reported by the Journal, Simo, who leads development of the new app, said the company was spreading its “efforts across too many apps and stacks, and that we need to simplify our efforts.”

ChatGPT is the signature chatbot from OpenAI, Codex is a coding platform designed for software developers, and Atlas is the AI-first browser from the company, which acts like a traditional internet browser, but with ChatGPT as an assistant.

(Disclosure: Ziff Davis, parent company of CNET, in 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

By creating a single app, OpenAI hopes to better compete with rivals like Anthropic. Responding to the Journal report in a post on X, Simo said the move is intended to build on the recent success of Codex, a competitor to Anthropic’s Claude Code.

“Companies go through phases of exploration and phases of refocus; both are critical,” Simo said. “But when new bets start to work, like we’re seeing now with Codex, it’s very important to double down on them and avoid distractions. Really glad we’re seizing this moment.”

A representative for OpenAI did not immediately respond to a request for comment.

Earlier this week, OpenAI announced its new GPT-5.4 mini and nano, smaller and faster versions of its ChatGPT 5.4 model. These coding models also highlight the company’s focus on supporting coders and enterprises instead of dabbling in various projects.

If you’re looking for an upgrade to your home audio, but you don’t have a lot of room to spare, the Bose Smart Soundbar is an excellent option. Normally $499, it’s currently marked down to just $419 on Amazon. With full Dolby Atmos spatial audio, a unique musical profile, and all the features you’d expect from a modern soundbar in a small package, there’s a reason it’s our favorite compact soundbar.

The biggest selling point here is the design. The Bose Smart Soundbar has an impressively small footprint, perfect for apartment-sized entertainment centers and cramped living rooms. It’s just 2.2 inches tall, making it easy to slide under basically any screen, and its 27-inch width even makes it a viable option for smaller panels. There are a surprising number of speakers inside, including a pair of proper up-firing drivers, so you get real spatial audio and full Dolby Atmos support, something fairly uncommon for soundbars.

Despite the size, the Bose Smart Soundbar has a great audio profile and feature set that’s just as good as any full-size bar. It has a more musical quality than most, which works just as well for an action movie or catching up on your favorite show as it does for listening to some music while you hang out on the couch. If you have a pair of Bose Ultra Open Earbuds, you can even pair them up for a personal surround setup, a feature unique to the Bose.

If you often have trouble catching what characters are saying, there’s an AI Dialogue mode that boosts the clarity and volume of speech. It’s a feature we’ve seen on other soundbars, but it stands out here with its excellent implementation, bringing any dialog into sharp focus at the push of a button. You can really hear the difference when switching the feature on and off.

If you have a little more space under your TV, or you’re just curious what other options are out there, make sure to check out our full guide to the best soundbars. Otherwise, you can head to Amazon to grab the Bose Smart Soundbar for just $419.

The UK challenger bank is rolling out Starling Assistant to personal account holders today, billing it as the UK’s first agentic AI financial assistant. It can set up savings goals, organise bill payments, and even quiz you on your own spending, all from a voice or text prompt.

Starling Bank has been building methodically towards this moment for the better part of a year. In June 2025 it became the first UK bank to put a natural language AI interface over customer spending data, with Spending Intelligence.

In October it launched Scam Intelligence, which lets customers upload images of marketplace listings or suspicious messages and receive a fraud-risk assessment. Both ran on Google Gemini on Google Cloud. Both were billed as UK firsts. On Friday, the bank announced the next step: Starling Assistant, which it is calling the UK’s first agentic AI financial assistant.

The distinction matters, at least in principle. Spending Intelligence and Scam Intelligence are read-and-respond tools: they analyse data and surface outputs.

Starling Assistant is designed to act, to take a voice or natural language prompt, and execute banking tasks directly on the customer’s behalf. It is also the umbrella under which those earlier tools now sit, creating a single conversational interface for what was previously a set of separate features.

The range of what it can do is more specific than most agentic banking announcements tend to offer. A customer planning a holiday can say they need to save £500 for a trip to Paris in July and ask the assistant to calculate a monthly savings schedule and set up automatic transfers to a dedicated Space.

Someone wanting to restructure their finances on payday can instruct it to create Spaces for groceries, bills, travel, and eating out, and specify how much to route to each. The press release also lists the ability to answer questions about direct debits and outstanding bills, analyse transaction history with specific payees, and, a detail that stands out, generate a quiz in which the assistant asks the customer to guess their top merchant for the last month or the category where they spend the most.

One caveat worth flagging: voice prompts are enabled by the user’s mobile keyboard rather than by native voice recognition built into the assistant itself. The distinction is small in practice but matters for anyone expecting a fully hands-free experience.

The assistant also has an explicit welfare dimension. Customers with vulnerability or accessibility needs can ask for specialist support without speaking to a human agent: it can help hard-of-hearing customers set up Starling’s sign language service, guide customers through setting up gambling blocks, and point those in financial distress to specialist resources.

Harriet Rees, Starling’s group chief information officer, described the launch as the culmination of eight years of AI development at the bank. Raman Bhatia, the group chief executive, called agentic AI “the next step in banking.”

Starling Assistant is live for personal current account customers from today, with business and joint accounts to follow. It is opt-in, and Starling says all data remains within its Google Cloud environment and is not used to train the underlying models. Graham Drury, director of FSI at Google Cloud UK and Ireland, described the shift as moving from navigating complex app menus toward simply having a conversation about your money.

The context around this launch is not entirely uncomplicated. The bank was fined £29 million by the Financial Conduct Authority in October 2024 for anti-money laundering and sanctions screening failures covering the period from 2017 to 2023, a fine reduced from £41 million after Starling cooperated with the investigation.

Building a reputation as the UK’s most ambitious AI-driven bank requires that the underlying compliance infrastructure is demonstrably sound. The welfare features built into Starling Assistant, the opt-in model, and the explicit commitment not to use customer data for model training are all consistent with a bank that has spent the past year investing heavily in rebuilding regulatory credibility alongside its product roadmap.

In the neobank sector more broadly, the race to agentic AI is accelerating. Revolut has signalled it is exploring AI agents but has not yet launched a comparable product in the UK. Bunq launched an AI assistant in 2024. Klarna has deployed AI extensively across customer service. For the moment, Starling is making the argument that the field in UK retail banking remains its to define.