10 september 2026 · 14:34
OpenAI Navier-Stokes claim: what it means for AI
OpenAI claims an internal model has resolved the Navier-Stokes Millennium Prize problem, using ten thousand coordinating AI agents running in parallel for nearly four days. The announcement has already drawn sharp pushback from mathematicians, raising urgent questions about how science validates machine-generated proofs. This episode also covers Google's €13 billion Finnish data centre commitment, Microsoft's record Patch Tuesday, Apple's iPhone 18 Pro launch, and two free tools worth adding to your workflow today.
Beluister deze aflevering:
Transcript
Samantha: Welcome to The State of Tech, The European Edition, Thursday September ten, 2026. I'm Samantha Lawrence.
Bob: And I'm Bob Russell. Today: OpenAI claims a proof of one of maths' most famous unsolved problems, Google pours thirteen billion euros into Finnish data centres, Apple's iPhone 18 Pro arrives with a much more capable Siri, Microsoft ships its largest ever Patch Tuesday with two zero-days already under attack, and to close out, two things you can actually use today, the free open-source screen recorder ScreenToGif, and the read-later app Instapaper. Let's start with that OpenAI announcement, because if it holds up, it changes what we think AI can do.
OpenAI claims an internal AI model has resolved the Navier-Stokes Millennium Prize problem.
Samantha: OpenAI says an internal model, described as significantly more capable than its current flagship GPT-6 Astra, has produced a resolution to the Navier-Stokes existence and smoothness problem. That is one of the seven Millennium Prize Problems in mathematics, unsolved for a quarter of a century, and it deals with how fluids like water and air behave under the equations that describe them. According to OpenAI, roughly ten thousand coordinating AI agents worked in parallel for eighty-eight hours, followed by seventeen hours of formal verification using GPT-6 Astra itself.
Bob: And the company says it will not claim the one million dollar prize, which is an interesting choice. The announcement has already sparked a priority dispute, with other mathematicians alleging the proof was rushed and needs far more scrutiny before anyone accepts it. That is not unusual for a problem of this weight. Peer review for something at this level typically takes months, sometimes years. The Poincaré conjecture, the last Millennium problem solved, took the mathematical community about four years to verify after Grigori Perelman posted his work.
Samantha: What makes this different is the sheer scale of the setup. Ten thousand agents coordinating for days is not a single model thinking harder, it is a small digital research institute running in parallel. That is a template OpenAI clearly wants to showcase, whether or not this particular proof survives review. The bottom line: even if the maths turns out to be flawed, the operational demonstration is significant. It suggests these systems can now sustain long, structured reasoning across many collaborating instances.
Bob: The catch is that formal verification by another AI from the same family is not the same as independent human verification. That is the core of the pushback from mathematicians today. And it raises a broader question about how scientific claims get established when the work is largely machine-generated. Journals, universities, and funding bodies do not yet have settled processes for this. Expect that debate to intensify sharply over the coming weeks, especially as more labs attempt similar demonstrations.
Samantha: For Europe, this lands right in the middle of an ongoing conversation about AI safety frameworks and research investment. European labs and universities are watching closely, both for the scientific implications and for what it says about the compute gap. Running ten thousand coordinated agents for eighty-eight hours is not something most European research institutions can casually spin up. That reinforces existing calls in Brussels and national capitals for more sovereign compute capacity, and for AI safety rules that keep pace with capability jumps like this one.
Google commits thirteen billion euros to expand AI data centre infrastructure in Finland.
Bob: Google has announced a thirteen billion euro investment in Finland, roughly fifteen point one billion dollars, to expand its artificial intelligence and digital infrastructure. Ruth Porat of Alphabet made the announcement alongside Finnish Prime Minister Petteri Orpo. The money goes into the backbone needed to train and operate Google's growing suite of AI products. This is Google's largest single AI investment in Europe to date, and it is a clear signal about where the company sees the next phase of its European build-out.
Samantha: Finland has been quietly becoming a favourite for hyperscalers, and the reasons are practical. The cold climate cuts cooling costs dramatically, and the country has abundant renewable and nuclear power. Part of this deal includes a twenty-two year agreement covering half the output of Fortum's Loviisa nuclear plant. That kind of long-term energy contract is becoming essential, because AI training and inference are extremely power-hungry, and grid access is now the real bottleneck for data centre expansion across Europe.
Bob: What shifts here for the market is the concentration. When Google, Microsoft, and Amazon all pour tens of billions into a handful of Nordic locations, you get a European AI backbone that is geographically clustered and heavily dependent on a few energy deals. That creates jobs and tax revenue for Finland, but also raises questions about resilience, about grid capacity for other users, and about whether smaller European countries get left out of the infrastructure map entirely.
Samantha: The European angle is genuinely mixed. On one hand, this is significant foreign direct investment and cements Europe as a real hub for AI infrastructure. On the other, the underlying AI models trained on these servers are still overwhelmingly American. European policymakers have been pushing for more sovereign AI capability, and a giant Google data centre in Espoo does not automatically deliver that. The infrastructure is here, but the intellectual property and the model weights are not.
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.
Microsoft's record September Patch Tuesday fixes 964 flaws, including two zero-days under active attack.
Bob: Microsoft's September Patch Tuesday is the largest on record, addressing 964 vulnerabilities, with 104 rated critical. Two of those are Windows zero-day flaws that attackers are already exploiting to gain full system privileges on machines. The US Cybersecurity and Infrastructure Security Agency has added both to its known exploited vulnerabilities catalogue, and told federal agencies to apply the fixes by September twenty-two. For everyone else, the message is the same: patch now, not next week.
Samantha: Nine hundred and sixty-four is a striking number. To put it in context, a typical Patch Tuesday runs somewhere between sixty and one hundred and fifty fixes. Ten times that in a single month suggests either a huge backlog being cleared, a wave of newly discovered flaws, or both. The two actively exploited ones are the immediate concern, because a system-level privilege escalation is what attackers use to turn a small foothold into full control of a machine, and from there, into a network.
Bob: For European organisations, the practical steps are unglamorous but essential. Any Windows environment needs to have this month's update rolled out quickly, especially on servers and administrator workstations. For home users, this is a genuine reason to open Windows Update this evening and let it run. The two exploited flaws are the kind that ransomware crews and state-linked groups fold into their toolkits within days of disclosure. Delay costs real money and real data.
Samantha: There is also a broader signal here about the scale of the software security problem. When one vendor is shipping close to a thousand fixes in a single month, that tells you something about the complexity of modern operating systems and the pace at which flaws are being found. European regulators pushing the Cyber Resilience Act have pointed to exactly this dynamic. Products need to arrive more secure and stay supported longer, because the current cadence of patch-and-pray is starting to strain everyone.
Apple launches the iPhone 18 Pro with a more conversational Siri powered by Apple Intelligence.
Bob: Apple has launched its new flagship phones, the iPhone 18 Pro and iPhone 18 Pro Max, and the headline feature is a significantly more capable Siri. Apple describes it as more personal, more conversational, and much more useful, powered by Apple Intelligence in iOS 27. The hardware gets the usual annual bump, a new main camera with a variable aperture, better battery life, a new chip, and a redesigned cooling chamber. Pre-orders open September twelve, general availability September eighteen.
Samantha: The Siri upgrade is the story that matters, because Apple has been visibly behind on AI assistants compared to Google and OpenAI. If this version genuinely delivers on the promise of understanding context, holding a conversation, and taking action across apps, it closes a real gap. If it does not, Apple's reputation on AI takes another hit. Reviewers will find out quickly. The variable aperture camera is a nice-to-have for photography enthusiasts, but the assistant is what shapes how a billion people use their phones daily.
Bob: For European buyers, there is a specific wrinkle worth flagging. Apple has previously delayed some Apple Intelligence features in the European Union while it worked through Digital Markets Act compliance questions. Whether the full Siri upgrade lands in Europe on day one, or arrives later with adjustments, matters for anyone thinking of pre-ordering. Check the fine print before you buy, because feature availability on European devices has lagged the United States for the last two release cycles.
Samantha: The regulatory angle stays relevant. European privacy regulators will scrutinise how the new Siri handles personal data, what stays on device, and what goes to Apple's servers. Apple has built its brand around on-device processing and privacy, and it will be measured against that promise. The Digital Markets Act also raises questions about whether third-party assistants get equal footing on iPhone. That fight is not settled, and this launch reopens it for another year.
The free open-source tool ScreenToGif turns any screen recording into an easy shareable clip.
Bob: To close out, two things you can actually use today, and neither of them needs a subscription. First up: ScreenToGif. It is a free open-source Windows tool that records a portion of your screen and turns it directly into a lightweight animated clip you can drop into an email, a chat, or a document. No accounts, no cloud upload, no watermark. You pick the area, hit record, stop when you are done, and you have a shareable file in seconds.
Samantha: What makes it genuinely useful is the built-in editor. After you record, you can trim frames, add captions, draw arrows, or crop the image before you export. That means you can show a colleague exactly which button to click, or capture a bug that only shows up briefly, without needing full video editing software. Reviewers consistently praise how small and fast the app is. It launches instantly, uses very little memory, and does not try to upsell you into a paid tier.
Bob: It runs on Windows only, which is the main limitation. Mac users have alternatives, but if you are on a Windows laptop or desktop, this is a genuinely handy tool to have in your toolkit. Download it from the official site or through the Microsoft Store, both routes are free. The exported files are small enough to send through corporate email systems that normally block large video attachments, which is often where screen recordings fall down in practice.
Samantha: The typical use cases are teaching a colleague a workflow, reporting a software bug with visual evidence, or making a quick how-to for a family member who lives far away. It takes about two minutes to install and about thirty seconds to make your first clip. For anyone who regularly explains things over screen sharing or long emails, this can save real time. And because it is open source, there is no hidden data collection or telemetry to worry about.
The read-later app Instapaper quietly saves the web for the moments you actually have time to read.
Bob: Second thing worth trying today: Instapaper. It is the classic read-it-later app, and it is having a bit of a moment again as people push back against algorithmic feeds. The idea is simple. When you come across a long article you do not have time for right now, you save it to Instapaper with one click, and it strips out the ads, the pop-ups, and the clutter. Later, on your phone or tablet, you get a calm, book-like reading experience.
Samantha: The free tier covers most people's needs. You get unlimited article saves, clean reading across devices, and offline access, which is genuinely useful on trains or flights. There is a paid tier for extras like full-text search across everything you have saved, unlimited notes, and text-to-speech that reads articles aloud. The app is available on iPhone, Android, and any web browser, and it works well with browser extensions that add a save button next to your address bar.
Bob: What users often report is that Instapaper changes their relationship with the internet. Instead of doom-scrolling headlines, you build a small queue of things you actually chose to read, and you get through them at a pace that suits you. It also plays nicely with newsletters. Many people forward long newsletters into Instapaper so they read them in the same clean format as everything else, rather than fighting formatting inside their inbox.
Samantha: For European users, it is worth noting the app is straightforward on privacy, and the reading experience works entirely without needing a live internet connection once articles are downloaded. Setup takes about five minutes: create an account, install the browser extension, install the phone app, and you are ready. It is one of those tools that is not flashy, but the people who use it tend to keep using it for years.
Samantha: Today we covered: OpenAI's claimed proof of the Navier-Stokes problem, Google's thirteen billion euro Finnish AI investment, Microsoft's record Patch Tuesday with two active zero-days, Apple's iPhone 18 Pro with the new Siri, and two things to try yourself: the free screen recorder ScreenToGif, and the read-later app Instapaper.
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.