China Silicon-28 mass production: quantum chips shift

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

Bob: And I'm Bob Russell. Today: Anthropic retires its older Claude models and points users to what's next, Morgan Stanley opens its trillion-dollar wealth platform to AI agents, China cracks mass production of the world's purest silicon for quantum chips, and Germany puts driverless army convoys on the road with Daimler Truck. Then two things you can actually use this afternoon: the password manager 1Password just bought a clever access tool called Apono, and Snap's new Specs glasses bring AI to your face. Let's start with Anthropic.

Anthropic retires Claude Sonnet 4 and Opus 4, pushing users toward a fresh 4.8 generation.

Samantha: Anthropic has officially pulled the plug on two of its older AI models, Claude Sonnet 4 and Claude Opus 4. As of yesterday, June fifteen, the company stopped accepting programming requests to those versions, which means any business that built an app or a workflow on top of them has to move. Anthropic frames this as routine lifecycle management, but for customers who plugged these models deep into their products, "routine" still means a migration project. And a new version, Sonnet 4.8, is expected very soon.

Bob: This is what model retirement at scale now looks like. A year ago, Sonnet 4 and Opus 4 were the flagships, the models that ChatGPT's biggest rival was using to win over law firms, banks and coding teams. Today they're switched off at the back end. For developers that's a hard deadline, you cannot just keep calling the old version because it works nicely with your prompts. You rewrite, you retest, and you hope the new model behaves the same way under pressure. The pace of this cycle is now closer to consumer electronics than to traditional enterprise software.

Samantha: And that pace has a real European dimension. A lot of European companies, especially in regulated sectors, only just finished their procurement and compliance checks for the previous generation. Banks, insurers, hospitals, they document which model version is in use, who approved it, what the risk assessment said. When the supplier retires that model, the paperwork has to be redone, the audit trail updated, and the AI Act obligations re-checked. It's a quiet tax on innovation that doesn't show up in the press release.

Bob: There's also a competitive angle. Anthropic is signalling that 4.8 is good enough to make the older line obsolete, not just optional. That puts pressure on OpenAI, Google and Mistral to keep the gap small. For the European cloud customers picking a long-term AI partner, the lesson from this week is simple: assume the model you buy today will be gone in twelve to eighteen months, and design your systems so swapping the engine doesn't break the car. The companies that build that flexibility in will save themselves a lot of grief the next time a vendor announces a retirement date.

Morgan Stanley becomes the first Wall Street bank to let outside AI agents into its trillion-dollar wealth platform.

Samantha: Morgan Stanley has done something no other major Wall Street bank has tried yet. It's opening up its stock administration platforms, ShareWorks and Equity Edge, so that external AI agents can plug straight in and pull data and insights. Until now, corporate clients managed their employee share schemes through human-facing dashboards. From this week, an autonomous AI tool built by the client can do that work directly, and the bank plans to roll this out to all thirty-four hundred administration clients by next year.

Bob: To put that in plain language, imagine a big company that runs share schemes for thousands of staff. Today, a finance team logs in, runs reports, checks who has vested, exports spreadsheets. Tomorrow, that company's own AI assistant logs in on their behalf, pulls the numbers, cross-checks them against payroll, and drafts the filing. The human steps in only to approve. Morgan Stanley is essentially saying: our software no longer has to be the front door, your AI can walk straight through.

Samantha: It's a meaningful shift because most banks have been extremely cautious here. Letting machines authenticate, query and act on financial data raises immediate questions about who is liable when something goes wrong. If a corporate client's AI agent misreads a vesting schedule and triggers an incorrect tax filing, where does the blame land? Morgan Stanley going first matters, because every European bank's risk committee will now look at this and ask whether they need to offer the same thing to stay relevant.

Bob: And for European supervisors this is exactly the scenario the AI Act and DORA were written for. Critical financial infrastructure, autonomous software making real decisions, and a cross-border data flow on top. Expect the European Central Bank and national regulators to study this carefully. The bottom line is, the era of AI just summarising your bank statements is ending. From this year on, AI will start actually using your bank.

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.

China mass-produces the world's purest silicon, unlocking a key building block for quantum computers.

Bob: Chinese scientists have achieved mass production of Silicon-28, an isotope of silicon enriched to over ninety-nine point nine nine percent purity. That sounds like a lab curiosity, but it's actually a serious milestone in the global race to build working quantum computers. Silicon-28 has zero nuclear spin, which in practical terms means it doesn't interfere with the fragile quantum states that these machines rely on. Less noise means qubits that hold information longer, which means quantum systems that can actually do useful work.

Samantha: And the strategic point here is independence. Until now, ultra-pure Silicon-28 was a niche material with very few suppliers worldwide. Researchers in Europe, the US and Asia had to compete for tiny quantities. By cracking mass production at home, China removes a bottleneck for its own quantum programme and potentially becomes a supplier to others, the way it already dominates rare-earth processing. That changes the geopolitics of quantum hardware, not just the science.

Bob: For Europe this is a mixed signal. On one hand, more global supply of high-purity silicon could accelerate research in the Netherlands, Germany and France, where there are strong quantum programmes at places like Delft and Munich. On the other hand, becoming dependent on Chinese-made foundational materials right when Europe is trying to reduce strategic dependencies is exactly the situation the EU Chips Act was designed to avoid.

Samantha: It's also a reminder that quantum is moving from hype to industrial reality. Five years ago, we talked about quantum mostly in research papers. Now we're talking about isotope factories, supply chains and export controls. The hardware layer is getting real, and the country that controls the inputs will shape the field. Expect Brussels to add Silicon-28 to its list of critical materials before the year is out.

Germany puts driverless army truck convoys on the road with Quantum Systems and Daimler Truck.

Bob: The German armed forces have commissioned drone-maker Quantum Systems and Daimler Truck to build autonomous military convoys. The project is called InterRoC, and the goal is straightforward: take Daimler Truck's military vehicle platforms, bolt on Quantum Systems' autonomy and AI stack, and run unmanned supply convoys through tough terrain. Germany has been very open about why, the Bundeswehr is short of personnel, and logistics convoys are both labour-intensive and dangerous.

Samantha: The civilian parallel is obvious. The same self-driving technology that companies like Waymo and Aurora have spent years testing on American highways is now being repurposed for defence logistics in Europe. And honestly, military logistics in some ways is an easier problem, you control the route, you control the schedule, and you accept that the convoy may need to go off-road. A supply truck following a lead vehicle through a forest track is technically simpler than a robotaxi navigating central Berlin.

Bob: There's a broader European story here too. Defence spending across the continent has surged since 2022, and a serious chunk of that money is now flowing into dual-use technology, things that have both military and civilian applications. Quantum Systems started life making reconnaissance drones for surveying and agriculture. Daimler Truck builds trucks for supermarket chains. Put them together on a Bundeswehr contract, and you get a model that other European armies, and NATO, will watch closely.

Samantha: For European citizens, this also raises uncomfortable questions worth naming. Autonomous military vehicles open debates about accountability and the ethics of removing humans from the loop. The German project is framed as logistics only, no weapons systems, but the line between a self-driving supply truck and a self-driving combat vehicle is technical, not philosophical. Expect this to become a real policy conversation in Brussels and in national parliaments over the next year.

1Password buys Apono to kill the dormant access rights lurking inside your company.

Bob: Now to two things you can actually try or look at today. First up, the password manager 1Password has acquired a smaller company called Apono. And while this is technically a corporate deal, the tool behind it is genuinely useful for anyone who works in or around IT. Apono does something called just-in-time access governance. In plain language, it makes sure people, machines, and AI agents only get permission to do something at the exact moment they need it, and that the permission disappears automatically when the job is done.

Samantha: That matters because one of the biggest invisible security risks in any organisation is what experts call dormant permissions. Someone gets access to a sensitive system for a one-off project, the project ends, but the access stays. Multiply that by hundreds of staff and dozens of AI tools, and you've got a quietly growing attack surface. Apono's approach is to flip the default, no permanent access at all, only temporary access tied to a specific task, then revoked.

Bob: For listeners who run a small business or manage a team, the practical takeaway is this: check out 1Password's site this week, where the combined product is being positioned as a single control panel for who and what can do what, across your apps and AI tools. Even if you don't buy anything, the concept is worth borrowing. Go through your own software and ask, who still has admin rights they no longer need? You'll be surprised.

Samantha: It's also a useful preview of where security is heading. As more companies deploy AI agents that act on their behalf, the question of what those agents are allowed to do, and for how long, becomes central. Tools like Apono are betting that the answer is "as little as possible, for as short as possible." Security people call it the principle of least privilege, and for AI agents that principle becomes even more important than it already was for humans.

Snap's new Specs glasses bring built-in AI to your face, with privacy as a selling point.

Bob: To finish, something you can go and look at right now. Snap, the company behind Snapchat, has unveiled the sixth generation of its AR glasses, called Specs. These are consumer glasses, not a developer kit, and the headline feature is built-in AI that understands what you're seeing and where you are, and uses that to help with everyday tasks. Think of looking at a restaurant menu in a foreign language and getting an instant translation in your field of view.

Samantha: What makes this interesting compared to Meta's Ray-Ban glasses or Apple's Vision Pro is the positioning. Snap is leaning hard into privacy, emphasising that processing happens in ways designed to keep your surroundings and your conversations from being uploaded wholesale. Whether that holds up under scrutiny is another matter, but the marketing message is clearly aimed at users who want the convenience of always-on AI without the constant data extraction.

Bob: For European listeners, the practical question is when and how you can get them. Snap announced Specs at the Augmented World Expo today, and pricing and EU launch details are still being clarified. But if you're curious, head to Snap's Specs page on their website to see what the glasses do and sign up for updates. Reviewers describe the device as lighter than earlier AR glasses, which has been the main barrier for everyday wear.

Samantha: And the broader point is that AR glasses are finally crossing from prototype to product. Meta has its Ray-Bans, Apple has Vision Pro, Google is back in the game, and now Snap is making a serious consumer push. For Europe, this is the year regulators will need to grapple with what it means to have cameras and microphones on millions of faces in public spaces, and what the GDPR really means when your glasses are silently watching the world around you.

Samantha: Today we covered: Anthropic retiring its older Claude models, Morgan Stanley opening its wealth platform to AI agents, China's breakthrough in ultra-pure silicon for quantum chips, Germany's driverless military convoys with Daimler Truck, and two things to try yourself: 1Password's new Apono access tool, and Snap's new Specs AR glasses.

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.