AI Act Transparency Rules: EU Consultation Opens Now

Samantha: Welcome to The State of Tech, The European Edition, Sunday June twenty-eighth, 2026. I'm Samantha Lawrence.

Bob: And I'm Bob Russell. Today: the US government greenlights a partial release of Anthropic's most powerful AI model, ON Semiconductor swallows Synaptics in a major chip deal, the European Commission opens consultation on the AI Act's transparency rules, the financial reality check hitting AI companies burning billions a month, and to close out, two things you can actually use today: a free photo organiser called digiKam, and a privacy-focused weather app called Breezy Weather. Let's get into it.

Washington lets Anthropic release its Claude Mythos 5 model to a select group of US cybersecurity firms.

Samantha: We start in Washington, where the US government has given Anthropic permission to release its most powerful AI model, Claude Mythos 5, but only to a hand-picked group of American cybersecurity companies. This is a partial unlock. The Trump administration had previously blocked access to Mythos 5 entirely, and now it's letting trusted defenders use it, while keeping it off the open market. Anthropic says it's also still negotiating with the government to widen availability, both for Mythos 5 and for its other top-tier model, Fable 5.

Bob: What's striking here is the precedent. The US is effectively building a tiered approval system for frontier AI, where the most capable models are vetted before release and then handed out to specific industries first. Cybersecurity is the test case, because the argument is that the same model that could help attackers can also help defenders, and you'd rather have the defenders armed first. But the mechanism itself is new. A government deciding which companies get to use which version of a private AI model is a meaningful shift in how this technology reaches the market.

Samantha: And it puts Anthropic in an awkward spot. The relationship with the administration has been bumpy, and now the company is balancing commercial pressure to ship with political pressure to wait. Every week Mythos 5 is locked down is a week competitors can catch up.

Bob: For Europe, this matters on two levels. First, European cybersecurity firms are not on the trusted list, at least not yet, which means American defenders get a head start with the strongest tools while European ones work with older versions. Second, Brussels is watching the model closely. The EU AI Act focuses on risk categories and transparency, but the US is showing a different lever: pre-release gatekeeping by government. Expect European regulators to study whether something similar could be layered onto the AI Act for the most powerful systems, especially in sectors like defence, critical infrastructure, and health. Industry groups in Europe are already arguing that without access to the same frontier tools, European cyber defence falls behind by default.

ON Semiconductor agrees to buy Synaptics in a major chip deal aimed at edge computing.

Samantha: Over to a big chip deal. ON Semiconductor has agreed to acquire Synaptics, and this is a serious bet on what's called edge computing, which simply means doing the smart work on the device itself rather than sending everything to a data centre in the cloud. ON Semiconductor is strong in power management and sensors. Synaptics brings touch, display, and on-device intelligence. Put them together and you get the building blocks for cars, factory machines, and connected home devices that can think locally, react faster, and use less energy.

Bob: The logic is straightforward. Sending every sensor reading from a car or a factory robot to the cloud and back is slow, expensive, and risky if the connection drops. Doing it on the chip itself, right where the data is generated, is faster and more private. For European industry, this is directly relevant. German carmakers, Italian factory automation specialists, and Dutch chip equipment suppliers all sit in this supply chain. A consolidated American competitor with stronger edge capabilities raises the pressure on European players like Infineon, STMicroelectronics, and NXP.

Samantha: It also fits a wider pattern. After years of every conversation being about the giant chips that train AI models in massive data centres, the spotlight is shifting to the small chips that run AI in everyday objects. Brussels has been pushing the European Chips Act precisely to keep manufacturing capacity on the continent. A deal like this reminds everyone why that matters. If the next decade of growth is in smart vehicles and industrial automation, the country or region that owns the edge chips owns the margin.

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.

The European Commission opens consultation on transparency rules under Article 50 of the AI Act.

Bob: Brussels has opened a public consultation on the draft guidelines for transparency under the AI Act, specifically Article 50. The rules kick in on August second, just over a month from now, and the consultation window is deliberately tight. The guidelines spell out what providers and deployers of AI systems actually have to do: tell users clearly when they're talking to an AI, label AI-generated content in a way machines can read, and disclose when an AI is involved in a decision that affects someone.

Samantha: This is one of the most consumer-facing parts of the AI Act. If you chat with a customer service bot, it has to tell you it's a bot. If you see a video that was generated or substantially altered by AI, there has to be a machine-readable marker built in. And if an AI helps decide whether you get a loan, a job interview, or an insurance quote, you have to be told. The consultation is the last chance for companies, civil society, and member states to push back on specifics before the rules apply.

Bob: The pressure points are predictable. Big platforms argue that machine-readable labels on AI content are technically messy, especially when content gets edited, re-uploaded, and screenshotted across services. Smaller AI providers worry about compliance costs eating into already thin margins. Consumer groups, on the other hand, say the draft is too soft and want clearer visible labels, not just hidden metadata that only other software can read.

Samantha: For Europeans, this is the moment the AI Act stops being theoretical. From August onwards, you should start noticing AI disclosures in places you didn't before, in chatbots, in recommendation systems, in generated images on social media. And for any company building or deploying AI in Europe, the next four weeks are about getting compliance ready, because enforcement powers come with the application date. Companies based outside the EU are caught too, as long as they offer their AI to European users.

AI companies are spending three dollars for every dollar they earn, raising sharp questions about profitability.

Bob: Now to the money story everyone in the AI industry is talking about. A new report shows that several of the biggest AI companies are losing more than a billion dollars a month, and across the sector, firms are spending around three dollars for every dollar they bring in. The headline figure: an estimated five trillion dollars is expected to flow into AI data centres in the United States alone over the coming years. That is not a typo. Five trillion.

Samantha: The pattern is unusual. Normally you build capacity in response to demand. In AI, the capacity is being built first, in the hope that demand will catch up. Financial researchers describe this as growing in reverse, and they're starting to ask uncomfortable questions. If revenue doesn't scale to match the spend, the writedowns will be brutal. And it's not just the AI labs. The chipmakers, the power companies, the construction firms building the data centres are all riding the same wave.

Bob: For European investors, the implications are sharp. Venture funds in Europe have been chasing AI valuations partly because their US counterparts are. If the American capital cycle turns, European AI startups that priced themselves against US comparables face a painful reset. The bottom line is that European investors should pressure-test their AI bets against scenarios where revenue grows slower than infrastructure spend.

Samantha: There's also a European policy angle. Brussels has been debating how much public money to put behind AI infrastructure, with figures floated for European data centre capacity and sovereign compute. Compared with the hundreds of billions flowing in the US, European numbers look modest. Industry groups argue that without serious European investment, the continent stays dependent on American infrastructure for its most strategic technology. The counter-argument is exactly today's story: if American AI economics are this fragile, copying the playbook at smaller scale could be the worst of both worlds.

digiKam is the free open-source photo organiser that handles decades of pictures on your own machine.

Bob: To close out, two things you can actually try this afternoon. The first is digiKam. If you've got tens of thousands of photos scattered across hard drives, old phones, and SD cards, and the idea of uploading them all to Google Photos or iCloud makes you uneasy, digiKam is built for you. It's a free, open-source photo manager that runs on your own computer, on Windows, Mac, and Linux. It reads pretty much every image format including raw files from cameras, sorts by date, location, faces, and tags, and never sends anything to a cloud.

Samantha: What makes digiKam stand out is how deep the tools go. It handles face recognition locally, so it can group photos of the same person without anyone else's servers seeing those faces. It supports keyword tagging, ratings, and colour labels for serious photographers, and it has a built-in editor for quick fixes like cropping and colour correction. Reviewers consistently praise it as one of the few open-source projects that genuinely competes with paid alternatives like Adobe Lightroom on organisation, though the interface takes a bit of getting used to.

Bob: For European listeners specifically, digiKam fits well with privacy preferences here. Your photos stay on your hard drive, no subscription, no account, no terms of service that change every six months. The project is community-developed and has been around for over twenty years, which matters for trust. Setup is straightforward: download from the official digiKam site, point it at a folder of photos, and let it index. The first scan on a big library can take a while, but after that it's quick. Users say the geotagging feature, which plots your photos on a map by location, is worth the install on its own.

Samantha: And it's a genuine alternative for anyone trying to leave a paid cloud service without losing the convenience of search and tagging.

Breezy Weather is the privacy-friendly weather app with no ads and no tracking.

Bob: The second one is Breezy Weather. It's a free weather app for Android that does something most weather apps don't: it leaves you alone. No ads, no tracking, no account required, no notifications begging you to come back. It pulls forecasts from multiple sources, including European weather services, and lets you choose which one you trust. The design is clean and the information is dense without being cluttered.

Samantha: The reason this matters is that most free weather apps are, quietly, advertising and data businesses. They harvest your location constantly, sell it to data brokers, and use the weather as the bait. Breezy Weather is open-source, community-maintained, and explicitly built to avoid all of that. Users praise the customisable widgets, the air quality readings, and the detailed hourly forecasts. It's available through the F-Droid app store and through Google Play, and it works across most of Europe with proper local data, not just the American sources many global apps rely on.

Samantha: It's a small switch with a real payoff: a weather app that just tells you the weather. Today we covered: Washington's partial unlock of Anthropic's Mythos 5, ON Semiconductor buying Synaptics in a big edge-chip deal, Brussels opening consultation on AI Act transparency rules, the financial reality check on AI companies burning billions, and two things to try yourself: digiKam for your photo library, and Breezy Weather for a clean, ad-free forecast.

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.