8 juli 2026 · 13:25
Frontier AI Cyberattacks: Banking Regulators Act
Global banking regulators have issued a joint warning that frontier AI models can now autonomously identify software vulnerabilities, build exploits and chain them into multi-stage cyberattacks on financial systems. For European banks and payment providers, this is no longer a theoretical IT risk: supervisors are signalling that AI-enabled attack scenarios will enter formal stress tests. This episode covers what the guidance means in practice, plus Samsung's paradoxical earnings shock for chip markets, Taiwan's cross-border green energy diplomacy, and the US quantum integration push called Project Triad.
Beluister deze aflevering:
Transcript
Samantha: Welcome to The State of Tech, The European Edition, Wednesday July eight, 2026. I'm Samantha Lawrence.
Bob: And I'm Bob Russell. Today: global banking regulators sound the alarm on frontier AI as a cyber weapon, chip stocks tumble after Samsung's earnings spook the market, Taiwan hunts for green power in Indonesia to feed its AI boom, the US launches a big push to fuse three branches of quantum tech into one system, plus two things you can actually use today, the open-source photo editor RawTherapee and the private European search engine Mojeek. Let's get into it.
Global banking regulators warn that frontier AI models can now launch multi-stage cyberattacks on financial systems with almost no human help.
Samantha: We start with a warning that landed with a real thud in banking circles overnight. Regulators around the world, including the central bank in the Philippines, have put out a joint memo saying the newest generation of AI systems, what they call frontier AI, is now capable enough to find software weaknesses on its own, work out how to exploit them, and then chain the whole attack together into multiple steps. In plain language: an AI that can spot the crack, pick the lock, and walk through the door, all without a human typing anything. Anthropic's most advanced model, Mythos, gets a specific mention. Cybersecurity researchers cited in the memo say it is unusually good at spotting systemic weak points in banking infrastructure, the kind of dependencies that connect one bank's plumbing to another's.
Bob: And what makes this different from previous cyber warnings is the scale. A human hacker can only be in one place at a time. An AI-driven attack can probe thousands of institutions simultaneously, adapt when it hits a defence, and keep going. Regulators are telling banks to do three things right now: map their full attack surface, so they actually know what an outsider can see, upgrade to AI-driven defensive tools that can react at machine speed, and rehearse incident response for scenarios where the attacker is essentially never tired and never sloppy. Adjacent to this, there is also reporting that Washington has been controlling which allies get access to the most capable frontier models, including Mythos, in the run-up to this week's NATO summit in Ankara. European allies are pushing for wider access to the strongest defensive versions, arguing that fragmented access makes joint cyber exercises almost impossible.
Samantha: For European banks and their customers, this matters in very practical ways. The European Central Bank and national supervisors are almost certainly going to echo this guidance in the coming weeks. Expect fresh stress tests focused on AI-enabled attack scenarios, and pressure on smaller banks and payment providers, who tend to lag on cyber budgets. The bottom line for ordinary customers: your bank's ability to detect fraud and outages depends increasingly on whether its defenders have AI as good as the attackers do. And that is now officially a supervisory concern, not just an IT department problem.
Chip stocks tumble as Samsung's record profit paradoxically triggers fresh doubts about AI infrastructure spending.
Bob: On to the markets, and a genuinely strange reaction. Samsung Electronics posted a preliminary operating profit of fifty-eight billion dollars for the second quarter, a nineteen-fold jump year-on-year. That is a record. And yet, US semiconductor stocks fell sharply on Tuesday. Intel and Applied Materials each dropped around ten percent, AMD around eight percent. The reasoning from analysts: Samsung's numbers are so extreme that investors are starting to ask whether this is a peak rather than a floor. If memory pricing is this strong now, the argument goes, the only way from here might be down, especially if the giant AI capital spending by Microsoft, Google, Meta, and Amazon starts to slow.
Samantha: The core question underneath all this is whether the hundreds of billions of dollars big tech is pouring into AI data centres will actually generate the returns that current stock valuations assume. Because right now, chip companies, data centre builders, and even the utilities supplying power are all priced as if the AI spending boom continues indefinitely. Tuesday's selloff suggests that consensus is starting to wobble. Analysts point out that even a modest slowdown in AI capital spending would ripple through the entire supply chain, from memory makers to networking gear to construction firms.
Bob: For European investors this matters directly. ASML in the Netherlands, ASM International, Infineon in Germany, STMicroelectronics, all of them ride the same wave. Pension funds and index trackers across Europe have quietly built up big exposure to AI infrastructure, often without savers realising it. If sentiment shifts, that shows up in retirement accounts. It also matters for European industrial policy, because the EU Chips Act is built on the assumption that demand for advanced semiconductors keeps rising steeply. If the AI capex cycle cools, subsidised fabs in Germany and France suddenly have a very different business case to justify.
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.
Taiwan's AI ambitions run headfirst into an energy crunch, and it is now shopping for green power in Indonesia.
Bob: Taiwan wants to be a global hub for artificial intelligence. It already makes most of the world's advanced chips through TSMC, and it is building out huge cloud and data centre capacity to match. The problem: all of that needs electricity, an enormous amount of it. National power consumption is now projected to grow two and a half percent every year for the next decade, driven mainly by AI data centres and chip factories. That may not sound dramatic, but compounded over ten years it is a step change. And Taiwan committed to net zero emissions by 2050, a target that is now looking very difficult.
Samantha: The country is boxed in geographically. It is small, densely populated, and does not have much space left for large solar or wind projects. It is also phasing out its nuclear reactors, which used to provide a big chunk of low-carbon baseload power. So Taipei is doing something unusual: green energy diplomacy. It is talking to Indonesia about importing renewable electricity, effectively outsourcing part of its clean power generation to a neighbour with more land and more sun. Details are still being worked out, but the direction is clear.
Bob: This is exactly the kind of trade-off European countries will face soon. Ireland already has moratoriums on new data centres in parts of the country because the grid cannot cope. The Netherlands, Germany, and the Nordics are all wrestling with the same maths: AI needs power, power needs to be clean, and clean power takes years to build. Watch for European governments to start looking at cross-border green energy deals of their own, potentially with North Africa for solar, or with Iceland for geothermal. The alternative is either missing climate targets or slowing down AI development, and neither is politically comfortable.
The US launches Project Triad, a bold effort to fuse quantum sensing, networking, and computing into one working system.
Samantha: For our fourth story, something genuinely different: quantum technology. The US National Science Foundation has announced Project Triad, and the goal is ambitious. Until now, quantum research has moved forward in three fairly separate lanes. Quantum sensing, which measures things like magnetic fields and time with extreme precision. Quantum networking, which sends information in ways that are physically impossible to eavesdrop on. And quantum computing, the one everyone hears about, which promises to solve certain problems dramatically faster than today's machines. Project Triad wants to weld all three together into one integrated system that actually does useful work.
Bob: The practical targets are healthcare, energy, manufacturing, and safety. Think medical scanners that can detect disease markers far earlier because the sensors are radically more sensitive. Think power grids that use quantum sensors to spot problems before they cascade into blackouts. Think secure communications between hospitals or between banks that cannot be intercepted, even in theory. The NSF frames this as moving quantum out of the laboratory and into industry. That is the hard part. Plenty of countries have impressive quantum research; very few have turned it into products you can buy.
Samantha: For Europe this is a real strategic moment. The EU has its Quantum Flagship programme and strong research groups in Germany, France, the Netherlands and Denmark. Companies like IQM, Pasqal and Quantinuum have European roots. But integrated systems, sensing plus networking plus computing working together in the field, is where the commercial edge will be won. If the US pulls ahead there, Europe risks doing world-class science while American firms sell the products built on it. Expect louder calls in Brussels for more coordinated European quantum funding, probably tied to defence and critical infrastructure, where quantum sensing and secure communications have obvious applications.
RawTherapee is the free open-source photo editor that gives you full professional raw processing without any subscription.
Bob: Right, to close out, two things you can actually use today. First one, for anyone who takes photographs, whether on a proper camera or a modern phone that saves raw files. It is called RawTherapee. It is a free, open-source photo editor that handles raw camera files, the unprocessed originals that give you the most flexibility to adjust exposure, colour, and detail. It is the sort of tool that photographers used to pay a monthly subscription for.
Samantha: What makes RawTherapee stand out is how much control it gives you. Reviewers consistently praise its highlight recovery, so if you overexposed a bright sky you can often bring detail back that looks lost. It has advanced noise reduction for photos taken in low light, and colour tools that let you fine-tune skin tones or landscape hues in a very precise way. It runs on Windows, Mac, and Linux, it is completely free, and there is no account required and nothing gets uploaded to a cloud. Your photos stay on your own machine.
Bob: It is not the friendliest app on first launch. The interface is packed with options, which can feel intimidating. But there are excellent free tutorial videos on YouTube, and the official documentation walks you through the basics. A good weekend project: pick ten holiday photos you always meant to fix, download RawTherapee, and work through them. Users report the results rival paid software once you get past the learning curve. And because it is open source, it keeps improving without ever asking for your credit card. Free, private, and genuinely powerful, a rare combination.
Mojeek is the independent European search engine that builds its own index and does not track you.
Samantha: Second find, and it is a non-AI one deliberately, because sometimes the answer to information overload is just a simpler tool. It is called Mojeek. It is a search engine, but unlike most alternatives to Google, it does not quietly repackage Google or Bing results in the background. Mojeek builds its own index of the web, from scratch, on its own servers in the United Kingdom.
Bob: The pitch is straightforward. No tracking of your searches. No profile built up about you. No adverts targeted based on what you looked for last week. And because the index is genuinely independent, the results are often different from what you get on Google, sometimes in useful ways. Users report finding older articles, smaller websites, and niche blogs that big search engines have quietly buried under commercial content. It is not going to replace Google for every search, especially for very local or very fresh news, but as a second opinion it is genuinely valuable.
Samantha: It is free to use in a browser at mojeek.com. You can also set it as your default search engine in Firefox, Chrome, Safari, or Edge, it takes about thirty seconds. There is a paid tier for people who want more search features, but the free version is enough for most users. Given the concerns we discussed earlier about AI, cybersecurity, and how much data flows across borders, having a European search engine that keeps your queries out of the advertising machine is a small but meaningful choice you can make this afternoon.
Bob: And that is our episode.
Samantha: Today we covered: global banking regulators warning about frontier AI as a cyber weapon, chip stocks tumbling after Samsung's record earnings triggered AI capex doubts, Taiwan looking to Indonesia for green power to feed its AI boom, the US launching Project Triad to fuse three branches of quantum tech, and two things to try yourself: RawTherapee for photo editing and Mojeek for private search.
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.