How the EU Fast-Tracked Client-Side Scanning for Encrypted Messages

Mia: I'm Mia.

Milo: And I'm Milo. This is HackerNews Daily, from the Bri podcast family. Today: the EU fast-tracks client-side scanning for encrypted chats, an AI tutor beats human TAs in a randomized physics trial, and Flipper Zero bets its future on a community app platform.

Mia: The EU Council just fast-tracked a measure to put client-side scanning inside encrypted messaging.

Milo: That sounds like the Chat Control proposal that's been debated for years. What actually changed now?

Mia: The Belgian presidency used a rarely invoked urgency procedure to skip a full vote and lock in the Council position.

Milo: So they skipped public debate and a majority vote on scanning private messages?

Mia: The measure requires platforms to scan all shared images, videos, and links before encryption, using AI detection.

Milo: That means even end-to-end encrypted chats would run scanning software on your device.

Mia: Critics call it a client-side backdoor because the scan happens before the message is encrypted.

Milo: And because it was fast-tracked, national parliaments had no chance to push back publicly.

Mia: The procedure also lets the presidency declare a position adopted without revealing which countries objected.

Milo: That's a transparency problem for a law that rewrites digital privacy across the EU.

Mia: The Council now has a formal negotiating mandate, and it moves to trilogue talks with the Parliament and Commission.

Milo: So the fight continues, but the Council just gave itself a stronger hand going into those final negotiations.

Mia: Another study just dropped that measured something pretty specific: an AI tutor in a real university physics course, with a randomized controlled trial.

Milo: Right, and the results are the kind of numbers you rarely see in education research without caveats.

Mia: The team at Dartmouth ran this inside a standard introductory physics class. Students were randomly split into two groups.

Milo: One group got active recitation sessions led by human teaching assistants. The other group got the AI tutor instead, during those same sessions.

Mia: The AI wasn't just a lecture bot. It adapted to each student's answers, asked follow-ups, and kept them explaining their reasoning out loud.

Milo: So instead of measuring whether students liked it, they measured learning directly with a physics concept inventory, the Force Concept Inventory.

Mia: The effect size ranged from about 0.7 up to 1.3 standard deviations, depending on which comparison you look at.

Milo: That's huge. In education research, even 0.3 is considered a meaningful gain. A 0.7 means the average AI-tutored student scored better than roughly seventy-five percent of the control group.

Mia: And the high end, 1.3, pushes that closer to ninety percent. These aren't lab results from a small pilot. It ran across hundreds of students in a live course.

Milo: What makes this worth paying attention to is the replacement they tested. They didn't compare AI to nothing. They compared it to trained human TAs running structured active-learning sessions.

Mia: Exactly. The AI matched or exceeded a form of teaching that is already considered high-quality and evidence-backed.

Milo: The paper also notes something practical. The human-TA version is hard to scale, especially at cash-strapped public universities.

Mia: If an AI tutor can produce equal or better learning at lower marginal cost, it changes the access equation for large introductory courses.

Milo: That still leaves a question, though. The effect sizes measure conceptual gain in one semester of physics. Does it stick, and does it work the same way in a humanities course where reasoning is less rule-bound?

Mia: The authors don't try to answer that here. They frame it as a proof-of-scale demonstration with effect sizes strong enough to justify larger, multi-discipline trials.

Mia: The Flipper Zero team just published a new development roadmap, and the most interesting part isn't a new hardware spec.

Milo: It’s that they’re doubling down on community contribution as the engine for adding protocols, radios, and even whole subsystem modules.

Mia: Right. They’re building an official SDK and an app repository so people can write and share add-ons without patching firmware by hand.

Milo: That’s a shift from “here’s a hacker multitool” to “here’s a platform where the community builds the long tail of capabilities.”

Mia: And it reuses their existing companion app as the distribution channel, which means updates don’t need a PC or a janky sideload flow.

Milo: The concrete consequence: new radio protocols or NFC tricks can arrive faster because a contributor writes a module, submits it, and users download it on-device.

Mia: That also changes the device’s identity. It stops being a fixed gadget and starts behaving more like an app phone for radio exploration.

Milo: But the roadmap is still early. The team is open about needing to stabilize the SDK and figure out curation and trust before the repo is wide open.

Mia: Right. So the takeaway is that Flipper Zero's future isn't a single killer feature. It's a bet that opening the software layer will make the hardware stay relevant much longer.

Mia: Let's stay with that. The team just published a new development roadmap, and the most interesting part isn't a new hardware spec.

Milo: Yeah, a flying umbrella that follows you around.

Mia: Right. The idea is a drone-style canopy that hovers above your head, tracking your movement to block rain or sunlight without you holding anything.

Milo: The claim I saw is that it uses sensors and computer vision to stay positioned directly above the user, adjusting in real time as you walk.

Mia: So the benefit is clear: hands-free weather protection that adapts to your path, unlike a fixed patio umbrella or even a handheld one.

Milo: The question is how it handles wind, battery life, and safety. A flying object that close to a person needs serious reliability.

Mia: Exactly. And the source didn't detail those specs, so the concept is out there, but practical deployment is still a big open question.

Mia: We just talked about a flying umbrella, and now there's a different kind of personal-risk finding that might hit closer to home. A new study says cannabis users face a substantially higher risk of heart attack, and the risk isn't just for older adults.

Mia: The researchers found that any cannabis use was linked to a higher chance of heart attack, regardless of age.

Milo: Even in people under fifty?

Mia: Yes — the association held even when they controlled for things like tobacco and other health conditions.

Mia: We just talked about a single risk factor, but now let's look at the bigger picture. Europe's climate is shifting in ways you can actually see across seven key charts.

Milo: Right, and the first thing that jumps out is warming. Europe is heating up faster than the global average, isn’t it?

Mia: It is. Since the late 1970s, land temperatures in Europe have risen about 0.5°C per decade. That’s roughly double the pace of the world overall.

Milo: So a hotter baseline means extremes hit harder. One chart shows heatwave days climbing steadily, especially in the south.

Mia: And it’s not just heat. Precipitation patterns are splitting Europe in two. The north is getting wetter, while the south keeps drying out.

Milo: That split shows up in river flows too. Some northern rivers are swelling, but southern water supplies are shrinking when they’re needed most. A long-term squeeze.

Mia: From climate data to the quieter side of computing. There's a story about keeping history alive, and it popped up in two places this week.

Milo: Yeah, one is what if you could run Windows 2000 on a processor that most people have never touched—a DEC Alpha?

Mia: Right. There’s a new fork of the es40 emulator that makes that a lot smoother. It’s not just nostalgia—Alpha was a radical RISC design that influenced modern chips, and this keeps a slice of that architecture genuinely usable today.

Milo: So you get both: a working Windows 2000 environment and the Alpha instruction set underneath, without hunting down rare hardware.

Mia: Exactly. And the other piece is “Starring the Computer,” a site that catalogs every computer shown in film and TV. It traces how on-screen machines shaped public perception of tech, decade by decade.

Milo: Together they’re two sides of the same thing—one preserves the machine you can touch, the other preserves how culture saw the machine.

Mia: What if you could open a full PCB design suite without installing anything?

Milo: Someone actually got KiCad running in the browser. It works through a WebAssembly port now.

Mia: So all the heavy schematic and layout computation happens locally in the browser sandbox, but you never have to touch an installer.

Milo: Exactly. And that means you can share a project link and the other person just clicks to inspect your board, no setup required.

Mia: That lowers the review barrier way more than exporting static Gerber views ever could.

Mia: Smart home devices are getting AI that acts on our behalf, ordering groceries or locking doors. But what happens when the AI's goals don't match ours?

Milo: Right, and this new threat model shows it’s not just a tech problem, it’s a social one too.

Mia: The paper argues we should treat the AI agent like a new member of the household, one that can be tricked or nudged into harmful actions.

Milo: So it’s less about a hacker cracking a password and more about someone manipulating the AI with a cleverly worded request?

Mia: Exactly. A visitor could say, “It’s too hot in here,” hoping the thermostat AI will act, even if they didn’t directly change it.

Mia: That threat model for household AI agents is a reminder that convenience always comes with a trust question. Thanks for listening today.

Milo: We'll be back with more stories soon. This is HackerNews Daily, from Bri.