EU child social media rules: age limits & safe by design

Samantha: Welcome to The State of Tech, The European Edition, Wednesday September sixteen, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Brussels is expected to unveil new rules limiting minors' access to social media; Microsoft agrees to strict AI guardrails for American schools; TSMC pumps the brakes on using AI for its most sensitive chip research; an Australian minister accuses a data centre giant of using AI to lobby the government; and to close out, two things you can actually use today, the encrypted European office suite CryptPad, and the open-source home inventory app Homebox. Let's start in Brussels.

The European Commission is set to unveil sweeping new rules on how children can access social media.

Samantha: Later today, during Ursula von der Leyen's State of the Union speech, the European Commission is expected to publish a legislative proposal on minors' access to social media. This is the moment the child safety debate stops being a discussion and starts becoming actual European law. The plan is expected to sit alongside the Digital Fairness Act, and it draws on advice from the Commission's Special Panel on child safety online. The direction is a developmental, age-tiered framework, with the strictest limits on children under thirteen.

Bob: The key phrase in the panel's advice is "safe by design." That means platforms would need to prove their services are genuinely safe for young users before those users get in, rather than fixing harms after the fact. According to the plans, the proposal also targets personalised advertising aimed at minors and the murkier corners of influencer marketing. For parents, that could mean fewer algorithmic rabbit holes for their kids. For platforms, it means engineering choices that today are optional could soon be legal requirements across the whole single market.

Samantha: And this is not happening in a vacuum. Several European countries have already moved on their own. France has pushed age verification, Spain has floated a minimum age of sixteen for social accounts, and Australia's under-sixteen ban has been watched closely in Brussels. What the Commission is proposing today would give Europe one common baseline instead of twenty-seven different national rules, which is exactly what large platforms have been quietly asking for, even if they will not love the specifics.

Bob: The bigger shift is philosophical. For years, online child safety in Europe leaned on parental controls and user reporting. This proposal, if it goes through in the shape the panel suggested, flips the burden onto the platforms themselves. Meta, TikTok, Snap and YouTube would need to show their design choices, from recommendation feeds to notification loops, are appropriate for a thirteen-year-old, not just a thirty-year-old.

Samantha: The bottom line is this is one of the most consequential pieces of tech regulation Europe will discuss this autumn, and the exact wording published later today will decide how far it really goes.

Microsoft agrees to strict AI guardrails and privacy standards for American classrooms.

Bob: Microsoft has signed a comprehensive agreement with the American Federation of Teachers, the second-largest teachers' union in the United States, on how its AI tools can be used in schools. The deal is unusually specific. Microsoft is barred from using student or educator data to train its AI systems, with a narrow exception for data used to protect students. It cannot sell the data it collects, use it for advertising, or feed it into product development. And crucially, the agreement bans AI companions designed to create emotional attachment or dependency in children.

Samantha: That last point is the one that jumps out. There has been a wave of consumer AI products marketed as friends or study buddies, and educators have been increasingly uneasy about children forming what feel like real relationships with a chatbot. By writing that ban directly into a vendor agreement, the union has effectively said: not in our classrooms. It also sets a template. Once one big union secures these terms, other unions and education ministries have a ready-made checklist to bring to the negotiating table.

Bob: And that matters for Europe. Microsoft is deeply embedded in European schools through Teams, Office and its Copilot AI tools. Education ministries in Germany, France, the Netherlands and the Nordics have all been wrestling with how to green-light AI in classrooms without opening the door to data harvesting. The concessions Microsoft has now made in the United States, no training on student data, no advertising use, no emotional-attachment companions, become a very hard floor to argue against in European procurement talks.

Samantha: What this shows is a maturing market. A year ago, schools were mostly asking whether to allow AI. Now the conversation is about the exact terms. And the terms are getting sharper.

Samantha: Quick interruption. If you listen to The State of Tech regularly, hit that like button and subscribe, that way you'll never miss an episode. Okay, moving on.

TSMC pumps the brakes on using AI for its most sensitive chip research.

Bob: The world's most important chipmaker is being unusually cautious about artificial intelligence. TSMC's co-chief operating officer, Y.J. Mii, has said the company is exercising prudence when deploying AI to process sensitive information, particularly in research and development. The concern is straightforward, data leakage and data corruption if AI tools are not carefully controlled. This from the company that makes the advanced chips that power almost every serious AI system in the world.

Samantha: And Mii went further than a generic caution. TSMC is happy to use AI where the ground is well understood, for example in designing the automation software that helps lay out chips. But for developing the next generation of process technology, where, in his words, there are "many unknown" challenges, the company is deliberately staying conservative. He referenced the recent calls from Anthropic's Dario Amodei and OpenAI's Sam Altman for a slowdown in advanced AI development, which is a striking thing for a TSMC executive to align with publicly.

Bob: The implicit worry is industrial espionage at machine speed. TSMC's process recipes are among the most valuable trade secrets on the planet. If engineers paste sensitive information into an AI assistant that then stores, learns from, or leaks it, the damage would be almost impossible to reverse. So TSMC is picking its spots. Notable is that this is a supplier discipline story, not a policy story, TSMC is setting the tone for its own R and D floor, and its customers will feel it.

Samantha: For European chip designers and equipment makers, from ASML down to smaller specialists, TSMC's stance sends a signal. If the biggest foundry in the world is treating generative AI as high-risk for its crown jewels, expect similar caution to spread through the European semiconductor supply chain, especially in research divisions handling export-controlled technology.

Bob: And it complicates the industry narrative that AI is a universal productivity boost. In some corners of tech, the honest answer right now is: not yet, and not here.

An Australian minister accuses data centre giant NextDC of using AI in its lobbying emails.

Samantha: Over to Australia, where a genuinely modern political row has broken out. Victoria's Jobs Minister, Melissa Horne, has accused NextDC, one of the country's largest data centre operators, of using artificial intelligence to write correspondence lobbying her office. The accusation, reported this week, is based on alleged AI-created errors spotted in emails between the minister and NextDC's chief executive, Craig Scroggie.

Bob: The backdrop is a fight over a proposed ten-hectare expansion of what NextDC calls a hyperscale AI factory in Footscray, a suburb of Melbourne. Local residents have been complaining loudly about the sheer scale of the site and its proximity to housing. NextDC has been pushing for fast-tracked planning approval. And now the minister is essentially saying: your submissions read like they were written by a chatbot, and there are tell-tale mistakes to prove it.

Samantha: What makes this interesting is that it is not really about the errors themselves. It is about trust. Ministers and regulators receive thousands of lobbying letters. If they start suspecting those letters are AI-generated at scale, without human accountability, the whole channel of corporate-government dialogue starts to look different. Some jurisdictions may end up demanding disclosure when AI is used in official submissions.

Bob: And this lands right in Europe's lap. The AI Act does not specifically regulate corporate lobbying, but national transparency registers in Brussels, Berlin and Paris could easily adopt disclosure rules for AI-generated submissions. Data centres are also a live European topic, with Ireland, the Netherlands and Germany all wrestling with grid capacity and community pushback on hyperscale builds.

Samantha: For NextDC, the immediate risk is reputational rather than legal. But the broader pattern is clear. As AI writing tools become standard in corporate communications, the ability to spot them, and the political appetite to call them out, is growing in step.

Bob: To close out, two things you can actually use today, both European, both free, and both about keeping your own stuff in your own hands.

The encrypted European office suite CryptPad gives you a genuine private alternative to Google Docs and Microsoft 365.

Samantha: The first is CryptPad. It is a free, end-to-end encrypted office suite developed in France by the company XWiki, and it runs entirely in your browser. You get collaborative documents, spreadsheets, presentations, kanban boards, whiteboards and forms, all encrypted in a way that means even the server operator cannot read your content. For anyone who has ever hesitated before pasting a sensitive draft into a mainstream cloud editor, that is the whole point.

Bob: The setup takes about a minute. Go to cryptpad dot fr, no account needed for a quick test, and open a new document. You can share a link with colleagues and edit together in real time, much like you would in Google Docs. If you want more storage or a permanent workspace, a free registered account gives you that, and paid tiers or self-hosting are available for teams and organisations. The developers publish the code openly, and the project receives funding from European research programmes.

Samantha: Users often highlight it for three situations. Journalists working with sources, lawyers sharing draft contracts, and small teams inside larger organisations that do not want a specific project sitting on a US-hosted platform. It is not going to replace every feature of a full Microsoft 365 rollout, and complex spreadsheets are lighter than what accountants are used to. But for meeting notes, shared documents and quick collaborative planning, it is genuinely usable today.

Bob: One practical note. Because everything is encrypted with a key derived from the sharing link, if you lose the link you lose the document. So treat those links like passwords. Bookmark them, store them in a password manager, and do not paste them into a public chat. That trade-off is the price of the privacy model.

Samantha: A good afternoon test is to move one recurring meeting agenda into CryptPad for a week and see how the collaboration feels. Available on any modern browser, on Windows, Mac, Linux, iOS and Android.

The open-source app Homebox turns everything you own into a searchable inventory on your own machine.

Bob: The second one is Homebox, and it scratches a very ordinary itch. It is a free, open-source home inventory app that lets you catalogue everything you own, from kitchen appliances to camping gear to the cables in that one drawer everybody has. You can add photos, serial numbers, purchase dates, warranty details, receipts, and the location where each item lives. Then you can search across the whole lot in seconds.

Samantha: Two audiences tend to love it. People preparing for a house move, and people who have just had an insurance claim and realised they had no idea what was actually in their flat. Homebox is widely used by home-lab enthusiasts and small landlords, and reviewers praise it for being lightweight and refreshingly free of subscriptions or upsells. The whole project is under an open licence and hosted on GitHub.

Bob: The catch is that Homebox is self-hosted. You run it on your own hardware, most commonly a small home server, a spare mini PC, or a Raspberry Pi. If those words already sound intimidating, this is not the pick for you today. If you are comfortable running a container, setup is a single command and you are up and running in about ten minutes. Your data stays on your device, nothing leaves your network.

Samantha: A gentle way in is to start with one room. Photograph and log the ten most valuable items, add serial numbers and purchase dates, and see how it feels. Once you have that, adding the rest becomes almost meditative on a rainy Sunday. And the day something breaks, gets stolen, or needs a warranty claim, you have the paperwork in one place.

Bob: Homebox runs on Windows, Mac and Linux, and you access it through any browser on your phone or laptop. Look for the project called Homebox on GitHub, the one maintained by the community under the sysadminsmedia organisation.

Samantha: Today we covered the EU's expected new rules on minors and social media, Microsoft's AI guardrails for American schools, TSMC's cautious stance on AI in sensitive chip research, the row in Australia over AI-written lobbying emails at NextDC, and two things to try yourself: the encrypted European office suite CryptPad, and the open-source home inventory app Homebox.

Bob: Want to know more or react? Visit stateoftech.eu or email us at info@doorzetters.net.

Bob: State of Tech, the tech world in 15 minutes.