← The State of Tech RSS Feed

23 oktober 2026 · 13:55

Microsoft's diagnostic AI vs doctors, Taiwan chip strike

Microsoft says its diagnostic AI system MAI-DxO outperforms physicians on accuracy and cost in complex cases, according to research published in the journal Science. In this episode, Samantha Lawrence and Bob Russell assess what those results mean for European regulators, health budgets and the EU AI Act's rules for high-risk medical tools. The episode also covers Taiwan's first-ever semiconductor strike at Micron's Taoyuan plant and what it signals for AI memory supply in Europe.

AI Microsoft PHP Cybersecurity Anthropic Micron

Beluister deze aflevering:

Transcript

Samantha: Welcome to The State of Tech, The European Edition, Friday October twenty-three, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: Microsoft claims a diagnostic AI that beats doctors, critical PHP flaws exposing most of the web, Anthropic launches a proactive AI agent for professionals, Taiwan sees its first ever chip strike, plus two things you can try today: the free photo app Nitrux Pix and the open-source map tool JOSM. Let's start with Microsoft.

Microsoft publishes research claiming its diagnostic AI is nearly twice as accurate as human doctors.

Samantha: Microsoft has published research in the journal Science about a diagnostic AI system called MAI-DxO, and the headline numbers are striking. The system was one point seven times more accurate than physicians on common emergency department cases, and more than twice as accurate on the complex cases that get written up in The New England Journal of Medicine. On top of that, Microsoft says it did the job at up to two-thirds lower cost. The company is openly using the phrase medical superintelligence, which is a bold framing even by current AI standards.

Bob: What is interesting is how they tested it. Microsoft built a new benchmark called the Sequential Diagnosis Benchmark, which forces the AI to work through a case step by step, ordering tests and narrowing down the diagnosis, instead of just picking an answer from a multiple-choice list. That is much closer to how a real doctor actually works. Critics will rightly point out that a benchmark is not a hospital corridor, and that peer-reviewed clinical trials on real patients are a different bar entirely. But the gap in the numbers is large enough that regulators will have to engage with it.

Samantha: And the cost angle is what will catch health ministers' attention. European health systems are under serious financial pressure, with waiting lists in several countries at record levels. A tool that can help triage complex cases faster and cheaper is exactly the kind of thing budget holders dream about. The question is who carries the liability when an AI diagnosis is wrong, and how patients are told that a machine was involved in the first place.

Bob: There is also a workforce dimension. Doctors' unions across Europe have already pushed back on earlier AI tools, arguing they create more admin, not less. Microsoft's framing of superintelligence will not help that conversation. The more realistic path is probably as a second opinion for difficult cases, especially in rural hospitals where specialist access is thin. That is where the cost and accuracy claims would translate into actual patient benefit, assuming the results hold up outside the benchmark.

Samantha: And for the European Medicines Agency and national regulators, this adds pressure. The EU AI Act classifies medical diagnostic systems as high risk, meaning strict conformity assessments before deployment. Microsoft will need to clear that bar country by country, and hospital procurement cycles are famously slow. So even if the research is as strong as it looks, European patients should not expect this tool at their GP next year.

AI tools help uncover two critical PHP flaws affecting roughly seventy percent of the web.

Bob: Security researchers at Wiz have disclosed two critical unauthenticated remote code execution flaws in PHP, the programming language that still powers around seventy percent of websites. The vulnerabilities affect nearly every PHP version going back more than a decade. Emergency patches are now available, and administrators are being urged to update immediately. What makes this story newsworthy beyond the sheer scale is how the flaws were found: Wiz used advanced AI tools to help with the discovery, which is becoming a pattern in high-end security research.

Samantha: Unauthenticated remote code execution is about as bad as it gets. It means an attacker does not need a password, does not need to trick a user, they just send the right request and run their own code on your server. Combine that with PHP's footprint across WordPress sites, e-commerce shops, government portals and small business websites, and the attack surface is enormous. For European hosting providers and web agencies, this is a drop-everything moment: patch now, assume scanning for vulnerable systems is already underway.

Bob: The AI angle cuts both ways. Defenders used AI to find the bugs before criminals did, which is a win. But the same capability is available to attackers, who can now scan the internet for vulnerable servers faster and cheaper than ever. Wiz itself notes that AI significantly lowers the barrier to scaling attacks. We are entering a phase where the window between a patch being released and mass exploitation starting is measured in hours, not weeks.

Samantha: For European organisations, the practical to-do list is short but urgent. Check which PHP version your websites run, apply the vendor patches, and review server logs for unusual activity. Smaller companies that rely on an external web developer should be calling them today. National cybersecurity agencies in Germany, France and the Netherlands will almost certainly issue formal advisories, if they have not already.

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.

Anthropic launches Claude Duet, a proactive AI agent for US Pro and Max subscribers.

Bob: Anthropic has opened a waitlist for Claude Duet, a new agent product for its US Pro and Max subscribers, powered by a model Anthropic is calling Opus five point five. The pitch is different from a standard chatbot. Duet is designed to deliver morning briefings, work on tasks in the background between your messages, and only pop up to ask you something when a decision really matters. In other words, Anthropic is trying to make its AI feel less like a search box and more like a quiet assistant that gets on with things.

Samantha: The strategic framing matters here. Anthropic has spent the past year positioning itself as the serious enterprise alternative to OpenAI, with a strong focus on safety and workplace use. Duet pushes that further into agent territory, where the AI actually does things on your behalf rather than just answering questions. The approval-only-when-critical design is a direct response to what has been one of the biggest complaints about earlier agents: they either bother you constantly or do something foolish without asking.

Bob: The catch for European listeners is in the headline: this is a US-only launch to start. Anthropic says European users should expect similar features over time, but there is no date. Part of the reason is regulatory. Agentic AI that acts autonomously raises harder questions under the EU AI Act, around transparency, logging, and user control, than a plain chatbot does. Anthropic will want its compliance story airtight before rolling this out across the EU.

Samantha: For professional users in Europe watching from the sidelines, the signal is clear. The frontier is moving from chat to agents that work while you sleep. That changes how you think about subscriptions, data access, and what you are willing to let an AI touch inside your email or calendar. Even without Duet available locally, now is a sensible moment to set internal rules about what agents can and cannot do with company systems.

Taiwan sees its first ever semiconductor strike as Micron workers demand a share of AI memory profits.

Bob: Union members at Micron's Taoyuan fabrication plant in Taiwan have walked out, in what is being described as the country's first-ever semiconductor strike. Ninety-nine percent of the nearly two thousand voting members backed a twenty-four hour stoppage, which began on Thursday morning. The dispute is straightforward: workers want a bigger share of the profits generated by surging demand for AI-optimised memory chips, which have become one of the most lucrative products in the entire tech supply chain.

Samantha: Taiwan's chip industry has long had a reputation for industrial peace. Strikes simply did not happen at this scale, partly because of culture and partly because of how tightly the sector is woven into national economic strategy. That makes this walkout genuinely significant. High-bandwidth memory for AI accelerators is in short supply globally, with buyers reportedly booking capacity years in advance. Workers can see the revenue numbers Micron reports to Wall Street, and they want a visible share of it.

Bob: A twenty-four hour stoppage is not going to break the global AI supply chain by itself. But it sets a precedent. If Micron's Taiwan workforce can organise and get results, workers at other fabs across Taiwan and South Korea may follow. For European companies, from carmakers to cloud providers, who are already competing hard for memory allocations, any new source of supply friction is unwelcome. Memory pricing has been climbing for months, and labour action adds another variable.

Samantha: There is also a political dimension. The European Commission has been pushing to build more chip capacity on European soil through the Chips Act, partly to reduce dependence on Asian suppliers. Stories like this one strengthen the argument, though building a competitive memory industry in Europe from scratch is a decade-long project at best. In the meantime, European buyers remain exposed to whatever happens at a single plant in Taoyuan.

Nitrux Pix is the free open-source photo organiser that cleans up messy libraries on your own laptop.

Bob: To close out, two things you can actually try today, starting with Nitrux Pix. It is a free, open-source photo organiser aimed at people who have thousands of pictures scattered across folders, external drives, and old backups. The appeal is simple: everything runs locally on your own computer. Nothing is uploaded, nothing is scanned by a remote server, and there is no subscription waiting at the end. The app sorts images by date, lets you tag and group them, and strips out duplicates. For anyone sitting on fifteen years of digital photos, that alone is worth an afternoon.

Samantha: The context here is that most people have outsourced their photo memory to Google Photos, Apple's iCloud, or Amazon Photos, and are now either paying monthly or hitting storage limits. Nitrux Pix is part of a quietly growing wave of European and open-source tools pushing back on that model. Users consistently report that it handles very large libraries without slowing down, which is where some free alternatives fall apart. It runs on Linux-based systems primarily, so Windows and Mac users may need to check compatibility before diving in.

Bob: What makes it practical rather than theoretical is that it does not demand you commit to a whole new ecosystem. You point it at your existing folders, and it works with what is already there. If you decide a month later it is not for you, your photos are untouched in their original locations. That lower risk is often what tips people from thinking about leaving the big clouds to actually doing it. For European users specifically, keeping family pictures on hardware you own avoids any question about where that data physically lives.

Samantha: The one honest caveat is that if you want AI-powered face recognition and automatic album creation, the big commercial services are still ahead. Nitrux Pix is deliberately simpler. Think of it as a filing cabinet that works, rather than a magic assistant. For many people, especially those uncomfortable with facial recognition running over family photos in the first place, that trade-off is exactly the point. Download it this weekend, point it at your pictures folder, and see what it does.

JOSM is the free open-source editor that lets you fix the maps millions of apps rely on.

Bob: Second pick for today is JOSM, which stands for the Java OpenStreetMap editor. It is a free, open-source tool that lets you edit OpenStreetMap, the community-run world map that quietly underpins apps from cycling planners to delivery services. If a path near you is missing, a shop has moved, or a cycle lane has been added, JOSM is how volunteers get those changes into the map that everyone else then benefits from.

Samantha: The reason this matters is that a huge number of apps European listeners use every day, from hiking apps to car navigation to charity mapping projects, pull their underlying data from OpenStreetMap. When the map is wrong, every app downstream is wrong too. JOSM gives you proper tools: aerial imagery overlays, GPS track imports, and validation checks that catch mistakes before you publish. It is more serious than the beginner web editor, which makes it ideal for anyone who wants to contribute properly rather than dabble.

Bob: It runs on Windows, Mac and Linux, and it is genuinely free. The learning curve is real, so most newcomers spend their first hour watching a tutorial, but once it clicks, you can map your neighbourhood in an afternoon. For people who like the idea of giving something back to a European-rooted open project, this is a tangible way to do it. Pick a corner of your town, fix a few missing footpaths, and see your changes appear in apps across the continent within days.

Samantha: Today we covered: Microsoft's diagnostic AI claiming to beat doctors, critical PHP flaws uncovered with AI help, Anthropic's new Claude Duet agent for US subscribers, Taiwan's first semiconductor strike at Micron, and two things to try yourself: the free photo organiser Nitrux Pix and the open map editor JOSM.

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.