Claude Code Backdoor: China vs Anthropic, Europe Watches

Samantha: Welcome to The State of Tech, The European Edition, Friday July ten, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: China accuses Anthropic of hiding a backdoor in Claude Code, Micron pours three billion dollars into the American chip supply chain, US lawmakers move to lock down AI chatbots, Google claims a real breakthrough in quantum error correction, and to close out, two things you can actually use today: the encrypted messenger Session and the open-source home dashboard Homarr. Let's start with the story shaking up the AI world this morning.

China accuses Anthropic of hiding a backdoor inside its Claude Code assistant.

Samantha: A Chinese state-run cybersecurity regulator has issued a formal warning about what it calls a security backdoor inside Anthropic's coding assistant, Claude Code. The regulator, tied to China's Ministry of Industry and Information Technology, says specific versions of the tool contain a built-in monitoring mechanism that can send sensitive user data back without consent. Chinese authorities are urging immediate audits, and in some cases full uninstallation. Alibaba has reportedly banned its employees from using Claude Code starting today.

Bob: And Anthropic's response is worth paying attention to. The company doesn't deny the mechanism exists. It says it was an experimental security measure, aimed at catching account abuse and at stopping rival labs from copying its model behaviour, what the industry calls model distillation. So the technical facts aren't really in dispute. What's in dispute is the label. Is this legitimate anti-abuse telemetry, or is it surveillance dressed up as security? Depending on which side you ask, you get very different answers.

Samantha: And the timing matters. This is one of the first big public clashes where a Chinese regulator is treating a Western AI tool the way Western regulators have treated Chinese apps like TikTok and DeepSeek. The playbook is being copied, in reverse. For European companies, that's the uncomfortable part. Brussels has spent two years worrying about AI supply chain risk from Chinese models. Now the same argument is being used against an American lab, and European firms sit right in the middle.

Bob: For European users of Claude Code, and there are plenty in software teams across Berlin, Paris and Amsterdam, the practical question is what data actually leaves the machine when the assistant runs. Anthropic will now be under pressure to publish a lot more detail about that. And European regulators, particularly under the AI Act's transparency provisions, may well ask the same questions the Chinese regulator just asked, only in a much more procedural way. The bottom line is that trust in AI coding tools is no longer just a technical question, it's a geopolitical one, and European firms have to plan for both fronts at once.

Micron commits three billion dollars to rebuild America's chip supply chain at home.

Samantha: Micron, the American memory chip maker, is putting up to three billion dollars behind the US semiconductor supply chain. The headline number is big, but the interesting detail is where five hundred million of it is going: into a raw silicon wafer plant in Sherman, Texas, run by GlobalWafers. That's the very start of the chip production line, the blank discs on which everything else is built. Right now, most of that raw wafer supply comes from Asia.

Bob: And that's what makes this more than a press release. The demand story is AI, memory chips are one of the biggest bottlenecks in AI servers right now, but the strategy story is independence. Washington has spent years trying to make sure that if something happens in the Taiwan Strait, the American chip industry doesn't grind to a halt within weeks. Investing in raw wafers is one of the least glamorous parts of that, and one of the most important.

Samantha: For Europe, this is a mirror held up to a very uncomfortable question. The European Chips Act promised forty-three billion euros of public and private money to double Europe's share of global chip production by 2030. Two years in, progress has been slow, several flagship fabs have been delayed, and Europe still depends almost entirely on Asian wafer supply. Compared to the hundreds of billions flowing in the US and China, European commitments look modest, and industry groups have been arguing that the bloc needs to move faster on the earliest, dirtiest parts of the supply chain, not just the shiny final fabs.

Bob: There's also a jobs angle. Micron says the investment will create new roles and support local economies in Texas. Similar language was used around European chip projects in Dresden and Magdeburg, but those have faced cost overruns and, in some cases, cuts to promised subsidies. The bottom line: the global race for chip independence isn't slowing down, it's specialising, and Europe risks being left with the mid-layer while others lock down both ends.

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.

US lawmakers move to force AI chatbots to protect kids and be transparent with users.

Bob: Two US Representatives, Valerie Foushee and Greg Casar, have introduced a bill called the People-First Chatbot Act. It's aimed at putting federal rules around AI chatbots for the first time. The core ideas: no using minors' chat logs to train chatbots without consent, clear disclosure when you're talking to an AI and not a human, tight limits on targeted advertising based on chat content, and monthly safety assessments for the systems themselves. Users would also get the right to see and delete the data these chatbots hold on them.

Samantha: What makes this bill notable is that it's specifically about chatbots, not AI in general. That's a narrower, sharper target than the sprawling debates in Washington so far. The concern behind it is well-documented: young users forming intense relationships with AI companions, chatbots giving harmful advice, and a data economy quietly hoovering up some of the most intimate conversations people have. Whether the bill passes is another question, but it's a marker that US politicians are catching up to where European regulators have been for a while.

Bob: And that's the European angle. A lot of what's in this bill already exists in Europe under the AI Act and the GDPR. The right to access and delete data, disclosure that you're talking to a machine, restrictions on profiling minors, these are already part of the European legal framework, just spread across different rules. If the American bill does pass, in some form, it means European chatbot providers face a much more aligned regulatory landscape on both sides of the Atlantic, which actually makes compliance simpler for the bigger platforms. The bottom line: the transatlantic gap on AI rules is narrowing, and children's protection is the wedge that's closing it.

Google claims a real quantum computing breakthrough by teaching machines to correct their own errors.

Samantha: Google's Quantum AI team, working with Google DeepMind, is reporting a genuine step forward in quantum computing. The headline result is an unusually low logical error rate on what's called the surface code, one of the leading approaches to quantum error correction. In plain language: quantum computers are famously fragile, tiny disturbances cause them to make mistakes, and the whole field has been stuck on how to reliably clean those mistakes up. Google says it has found a better way.

Bob: The technique is clever. Instead of treating error correction and calibration as two separate jobs, they've merged them. The system uses the errors it detects as a training signal for a reinforcement learning agent, an AI that keeps tuning the quantum computer's controls while it's running. Think of it as a car that adjusts its own engine mid-drive based on the bumps it feels. Google's simulations suggest this approach doesn't slow down as the quantum machines get bigger, which has been one of the biggest doubts about scaling this technology up.

Samantha: For Europe, this matters in two ways. First, it puts even more distance between the very top of the quantum field, which is largely American, and the rest, including strong European players like IQM in Finland, Pasqal in France, and Quantinuum with UK ties. Second, it changes the timeline debate. Practical quantum computers have always been ten years away. Every time an error correction milestone like this lands, that horizon gets a bit shorter, and European industries from pharma to finance to defence have to start planning for it more seriously, especially on the encryption side.

Bob: And that encryption side is where it gets uncomfortable. The same quantum machines that could speed up drug discovery could eventually break some of the encryption protecting today's internet traffic. European governments have been pushing so-called post-quantum cryptography for a while, but progress in industry has been patchy. Announcements like today's are exactly the kind of nudge that moves that work up the priority list. The bottom line: quantum is still not here, but it's moving from science fiction into IT roadmap territory.

The private messenger Session hides your identity without any phone number or email.

Samantha: To close out, two things you can actually use today. The first is a messaging app called Session. It's a free, open-source messenger that positions itself as a step beyond apps like Signal on the privacy front. The difference: Session doesn't ask for a phone number or an email address to sign you up. Instead, you get a randomly generated ID. Messages are end-to-end encrypted, and they're routed through a decentralised network of servers rather than a single company's data centre.

Bob: The project is developed by a foundation now based in Switzerland, which is one of the reasons it's been picking up interest in European privacy circles. It runs on iPhone, Android, Windows, Mac and Linux, and there's a desktop version too. The trade-off is that because there's no phone number linking you to contacts, you have to share your Session ID or scan a QR code to start a conversation. It's less convenient than WhatsApp, but for journalists, activists, lawyers or anyone who simply doesn't want their metadata sitting in a big central database, that trade-off is the whole point.

Samantha: It's worth being clear-eyed about what it does and doesn't do. Session hides who you're talking to from the network itself. It doesn't stop the person on the other end from screenshotting your message, and it's not designed for huge group chats or business use. Reviewers generally praise the underlying design but note the interface feels slightly less polished than the big commercial messengers. If you've been curious about what a truly anonymous chat app looks like, this is one to try this weekend, download takes a couple of minutes and there's genuinely no signup form to fill in.

Bob: And the fact that it works across every major operating system means you can actually convince a few friends to join without anyone hitting a wall. That's usually where privacy apps fall down.

The free home dashboard Homarr turns any browser into your personal command centre.

Samantha: The second thing to try today is a tool called Homarr. It's a free, open-source dashboard that pulls all your favourite web services, home network tools, and bookmarks into one clean page you open in your browser. Think of it as a start page on steroids, one you host yourself, so nothing goes through a third party. It's aimed at people who juggle a lot of tabs, run a home server, or just want a tidy jumping-off point in the morning that isn't cluttered with ads.

Bob: You can add widgets for the weather, your calendar, your media library, network monitoring, even live status checks on the websites you rely on. Users describe it as the calmest way to see everything you care about at a glance, without a subscription and without a company tracking which links you click. The setup takes a bit of technical patience the first time, you install it on a home computer or a small server, but there's a friendly community and detailed guides, and once it's running, it just works.

Samantha: For anyone in Europe curious about self-hosting, running your own tools instead of relying on Google or Microsoft dashboards, Homarr is one of the gentler places to start. It's a small, satisfying weekend project, and it doubles as a nice introduction to keeping more of your digital life on your own hardware, which is a theme European privacy advocates have been pushing for years.

Samantha: Today we covered: China's backdoor accusation against Anthropic's Claude Code, Micron's three billion dollar bet on the American chip supply chain, the US chatbot safety bill catching up with Europe, Google's quantum error correction breakthrough, and two things to try yourself: the anonymous messenger Session and the self-hosted dashboard Homarr.

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.