US export controls forced Anthropic to kill two AI models globally overnight. Pokémon Go players unknowingly built military drone navigation from 30 billion street scans. Plus: Wi-Fi that recognises you by your walk, and flatworms that may eat memories. [DVYio Newsletter](https://newsletter.dvy.io/visit/uxdZ7Zm4) Tech [A giant hand holds a locked cage containing two glowing AI models in front of a world map and silhouetted people.](https://newsletter.dvy.io/visit/ajYwCEw3) [US government forces Anthropic to pull Fable 5 and Mythos 5 over a jailbreak concern](https://newsletter.dvy.io/visit/ajYwCEw3) Late on Friday, under a government export control order, Anthropic had to disable two of its AI models, Fable 5 and Mythos 5, for every customer worldwide. The directive prohibits any foreign national from accessing them, including Anthropic's own foreign national employees, so the only way to comply was to pull the models entirely. The stated reason is a jailbreak. The government says it has found a technique that bypasses Fable 5's safety controls and is concerned about the national security implications. Anthropic disputes the severity: their review of the specific technique found it exposed a small number of previously known, minor vulnerabilities, and they say other publicly available models can surface the same information without any bypass at all. Anthropic's broader position is that Fable 5 launched with the strongest safeguards of any model to date, validated by thousands of hours of red-teaming with the US and UK governments and multiple independent organisations. [Five-step diagram showing Wi-Fi beamforming data capturing a person’s walk, turning it into a gait signature, and identifying them with high confidence.](https://newsletter.dvy.io/visit/V5Bif3WS) [Your Wi-Fi router can identify you by how you walk](https://newsletter.dvy.io/visit/V5Bif3WS) Your WiFi router is already broadcasting a signal into the room. What this research shows is that signal can be used to identify you specifically, just from the way you walk. The mechanism is beamforming, a feature baked into WiFi 5 and everything since. To aim its signal efficiently, a router asks your device to report back what the wireless channel looks like from where you are. Your phone or laptop sends that report automatically. The researchers found those reports, called beamforming feedback information (BFI), contain enough detail about how a body is moving through space to act as a biometric. The attack, which they call BFId, was tested on 197 people. It identified individuals with high accuracy even when they varied their walking style or approached from different angles. No cooperation required from the target, no special hardware beyond a standard WiFi setup. BFI reports are just considered housekeeping data, not something people or device manufacturers think of as sensitive. Most privacy work on WiFi sensing has focused on things like motion detection or health monitoring. The identity inference angle is new, and there is no existing defence against it. [Apple OS 27 icon with a large white “27” over a glossy beige and grey abstract background.](https://newsletter.dvy.io/visit/SGu333pP) [Apple is rebooting Siri for iOS 27](https://newsletter.dvy.io/visit/SGu333pP) This week, Apple gave their annual preview of what will be arriving in the new operating systems across their devices. The big announcements were around Siri and AI. Siri has always been a tap-and-command assistant. You bark an instruction, it does (or fumbles) the one thing, and the interaction ends. Apple's OSes 27 changes that model across the board: iOS, iPadOS, macOS, watchOS, and visionOS (all arriving this autumn). The headline feature is "Siri AI", which is Apple's name for a conversational, context-aware Siri. Instead of each request being its own isolated task, Siri now holds the thread across a back-and-forth exchange. The focaccia example Apple uses is a decent illustration: you ask Siri to find a recipe "Sarah" emailed you, and then follow up with "add those ingredients to my shopping list" without repeating yourself. Siri remembers what you were just talking about. You can type or speak, whichever suits the moment. Siri AI launches in English only, with other languages coming at some unspecified point. Conversational AI is significantly harder to build well in languages other than English, and Apple has a history of rolling these features out slowly across regions. If your phone isn't set to English, the headline feature may not arrive for quite some time. And if you're in the EU, you won't get a big chunk of the features for now. [A group of Pokémon Go players pose around a large Pokémon Go Fest 2026 sign in front of a city fountain and skyline.](https://newsletter.dvy.io/visit/XuP9QgQK) [Players who scanned their streets for Pokémon Go rewards unknowingly trained military drone navigation](https://newsletter.dvy.io/visit/XuP9QgQK) Roughly 30 billion street-level scans collected by Pokémon Go players are now powering military drone navigation. Players filmed their surroundings between 2021 and the present to earn in-game rewards, each short video adding to a 3D model that lets a machine work out its position by sight when GPS is unavailable. Most had no idea that's what they were contributing to. The scans belong to Niantic Spatial, which in December 2025 announced a partnership with Vantor (formerly Maxar Intelligence, a US defence and intelligence contractor) to combine that ground-level visual navigation system with Vantor's aerial software for use in GPS-denied military operations. The terms players agreed to gave Niantic a transferable, sublicensable licence to the footage, so selling or licensing it to third parties was always legally possible. Whether players understood that at the time is a different question. Floris De Hingh, a Dutch player who'd been scanning locations since 2016, told Dutch newspaper Trouw he thought he was just playing a game. The training data pipeline here goes from "scan this bench to get a Poké Ball" to "navigate a military drone over a contested area", and the people who built that dataset weren't told where it was headed. [Apple promo image of an iPhone with the text “New Siri AI” and “0.1” on the screen, plus the date Wed Apr 1.](https://newsletter.dvy.io/visit/vCL5wTao) [A dark screen shows stacked Siri suggestion buttons for actions like create a reminder, call Vicki, rotate photo left, and send an email.](https://newsletter.dvy.io/visit/vCL5wTao) [An iPhone lock screen shows a Siri prompt with a text message from Aga about Calathea, described as a patterned tropical houseplant.](https://newsletter.dvy.io/visit/vCL5wTao) [A Pages document titled "Intern Onboarding" with a typed prompt asking for a quick guide to product photography for a summer intern.](https://newsletter.dvy.io/visit/vCL5wTao) [Apple's new Siri](https://newsletter.dvy.io/visit/vCL5wTao) This Week I Learnt [Infographic showing same-day letter delivery in Victorian London, with up to seven collections a day, fast transport, local sorting, 2-3 hour delivery, and delayed mail causing stale decisions.](https://newsletter.dvy.io/visit/ToSLLYhb) [Victorian London had same-day letter delivery, multiple times a day](https://newsletter.dvy.io/visit/ToSLLYhb) [Victorian London had up to seven postal deliveries a day within the city. Post a letter before 10am, and it would likely arrive across town before lunch. The system was so reliable that Londoners used it the way we use text messages: fire off a note in the morning, get a reply by the afternoon. Which makes the complaint from a Times reader in May 1881 feel very familiar. He'd posted a letter at half one from Gray's Inn, addressed to someone near Westminster Abbey (about a mile away), and it didn't arrive until 9 o'clock that evening. Eight hours. He wasn't angry that same-day delivery had failed; he was angry because the delay had real consequences. If everyone assumes a letter takes two or three hours and it actually takes eight, people make decisions based on information that's already out of date. We're not so different. We've just moved the expectation of immediacy from letters to email, then to Slack, then to WhatsApp. The underlying anxiety, that someone is acting on stale information because the message arrived late, is identical.](https://newsletter.dvy.io/visit/ToSLLYhb) Poll The Pokémon Go story: would knowing the data might be used for military drones change how you'd have played? Never would have scanned Still would have scanned Didn't play, doesn't apply Last Week's Poll What do you think actually drives most people's major life decisions? Personal values 19% Money 30% Relationships 6% Chance and circumstance 30% Social approval 15% Science [Diagram of Earth with the Sun on the left and dashed arrows showing sunlight reflecting symmetrically around the planet.](https://newsletter.dvy.io/visit/3J7BQZeW) [Earth reflects sunlight almost perfectly symmetrically — and nobody knows why](https://newsletter.dvy.io/visit/3J7BQZeW) Earth reflects sunlight back into space, and for decades we assumed the north and south hemispheres did this roughly equally. That was already strange, given how different they are (one is mostly ocean, the other mostly land). Now researchers at the University of Colorado Boulder have found a second, equally puzzling match: split the planet at the 27° East meridian and the eastern and western hemispheres also reflect almost identical amounts of sunlight. And apparently there's a third symmetry too, making it a "triple symmetry" in total. The continents are arranged nothing like a mirror image. The oceans aren't symmetrical. And yet the numbers balance out anyway, repeatedly, across different axes. However, that long-established north-south balance may now be breaking down as both hemispheres darken. Whether the east-west symmetry is similarly fragile isn't yet clear. [A stylised black shoe sole floats amid bright, wavy polymer chains and starburst shapes on a green background.](https://newsletter.dvy.io/visit/FH2jvZvz) [A new class of plastic that absorbs impacts by breaking apart at the molecular level](https://newsletter.dvy.io/visit/FH2jvZvz) Most plastics fail by cracking: stress builds up, a crack forms, and the material splits. MIT chemists have taken a different approach, designing a cross-linking molecule (think of it as a microscopic safety valve woven into the plastic's structure) that sacrifices itself under impact before the surrounding material can crack. The valve breaks, absorbs the energy, and the bulk of the plastic stays intact. The result is polystyrene and shoe-sole rubber that are roughly twice as strong as their standard versions. The cross-linker is the only thing that changes; the base plastic is the same. The analogy that comes to mind is a crumple zone in a car: you deliberately build in something that fails in a controlled way, so the thing you care about doesn't. [Surreal illustration of a scientist with blue gloves using a microscope to observe a bisected planarian flatworm.](https://newsletter.dvy.io/visit/cZ27tyZu) [Can flatworms pass memories to the worms that eat them?](https://newsletter.dvy.io/visit/cZ27tyZu) Sometime in the early 1960s, a researcher named James McConnell trained flatworms to flinch at a flash of light, then ground them up and fed them to untrained worms. Those untrained worms, the experiment suggested, picked up the conditioned behaviour faster than chance alone could explain. The implication was genuinely strange: that memory might be stored in transferable molecules, not just in the physical wiring of a brain. That you could, in some sense, eat a learned behaviour. The idea briefly captivated the scientific mainstream before the establishment turned on it. Replication attempts failed, the researchers scattered, and the whole episode got filed away as a cautionary tale about wishful thinking in neuroscience. Claire L. Evans traces what happened to the scientists who took it seriously anyway, and whether the question itself was ever properly put to rest. [Four-step diagram showing circularly polarised light writing a valley state in WS2, then emitting and routing chiral light through a Y-shaped waveguide to WSe2 detectors.](https://newsletter.dvy.io/visit/iiZwi8kj) [A chip that steers light using the quantum property of which valley an electron sits in](https://newsletter.dvy.io/visit/iiZwi8kj) Most computers encode information using electrons. Valleytronics tries a different angle: electrons in certain ultra-thin materials can sit in one of two distinct energy states (called valleys), and you can switch between them using light. Left-circularly polarised light targets one valley; right-circularly polarised light targets the other. The field has had a plumbing problem. Researchers could generate valley-dependent light, or route it, or detect it, but nobody had stitched all three together in one device. This paper does. A monolayer of tungsten disulfide converts incoming infrared light into visible light with a spin (left or right circular polarisation) that directly reflects which valley the electrons were in. A nanoscale waveguide with patterned structures sorts that light by handedness and routes it accordingly, with a polarisation selectivity of 0.97, close to perfect. A second ultra-thin material at the far end reads the light back out as an electrical signal. Generate, sort, detect, all on one chip, at room temperature. They demonstrated it by encoding and processing images through valley states simultaneously. It works. Valleytronics has been a phenomenon in search of a device for a while. This is a credible first one. [Thatched-roof studio with large glass walls, set among pine trees and tall grass.](https://newsletter.dvy.io/visit/ZkU49SDh) [Wood-lined studio interior with a long table, chair, and large windows looking onto dune grass and trees.](https://newsletter.dvy.io/visit/ZkU49SDh) [A thatched-roof wooden cabin with a partially open glass door, curtain, and deck overlooking dunes.](https://newsletter.dvy.io/visit/ZkU49SDh) [Wood-lined room with a brick floor, bench, and large windows framing grass and the sea beyond.](https://newsletter.dvy.io/visit/ZkU49SDh) [A small wooden studio with a steep shingled roof sits amid dense coastal greenery, with a glass corner facing the garden.](https://newsletter.dvy.io/visit/ZkU49SDh) [A thatched-roof wooden studio with glass doors opening onto a deck, set among dunes and trees.](https://newsletter.dvy.io/visit/ZkU49SDh) [Shoreline Studio](https://newsletter.dvy.io/visit/ZkU49SDh) [Odsherred, Denmark](https://newsletter.dvy.io/visit/ZkU49SDh) Data [A world map from Our World in Data shows the primary electricity source by country for 2024/25. Coal (brown) and gas (orange) dominate globally, particularly in Australia, China, India, and the US. Hydropower (blue) is the main source across much of South America, Canada, and parts of Africa, while oil (red) is concentrated in the Middle East and Northern Africa. France stands out as nuclear-reliant, and few countries, such as Spain and Chile, have solar or wind as their top generator.](https://newsletter.dvy.io/visit/2x9hLp3Y) [Mapped: the biggest source of electricity in every country](https://newsletter.dvy.io/visit/2x9hLp3Y) [Coal is the single largest source of electricity on the planet, responsible for about a third of global generation. That figure is so high largely because of Asia: China, India, Indonesia, and Malaysia all run primarily on coal, and they are enormous power producers. Zoom out to the rest of the world, though, and it gets more interesting. Gas and hydropower dominate across most of South America, Africa, and Central Asia. Oil is surprisingly prevalent on islands and parts of North Africa, where grid infrastructure limits the options. Europe is the most varied, with nuclear leading in France and Finland, and solar and wind now the biggest single source in Spain and Germany. The renewables picture is slightly flattering depending on how you count. Wind and solar are variable (they only generate when the sun shines or the wind blows), and they're often treated separately in energy statistics. If you combine them into a single "variable renewables" category, six more countries join the list where clean sources dominate: the UK, Netherlands, Portugal, Greece, Belgium, and Pakistan.](https://newsletter.dvy.io/visit/2x9hLp3Y) History [Collage of invasion-fiction images, centred on the cover of The Battle of Dorking, with soldiers, maps, and wartime scenes.](https://newsletter.dvy.io/visit/9pLEQcCs) [British invasion fiction: the forgotten genre that feared Europe before we did](https://newsletter.dvy.io/visit/9pLEQcCs) [There's a Victorian literary genre almost nobody reads any more: invasion fiction. From roughly 1871 to the outbreak of the First World War, British writers churned out hundreds of stories, pamphlets, and novellas imagining what would happen when a European power finally crossed the Channel and took Britain apart. The quality was, by most accounts, poor. These weren't prestige novels; they were closer to political pamphlets wearing a narrative disguise, often written to argue for higher military spending. But they had a formula that clearly worked on readers. Scholar A. Michael Matin catalogued more than a hundred of them and found ten motifs that appear so consistently they're basically the genre's DNA: near-future settings, subversive foreign spies on British soil, incompetent Liberal politicians, and a general sense that the public was sleepwalking toward catastrophe. What's interesting isn't the individual books but what the genre reveals. Britain's island geography, which most people would assume made it feel safe, apparently produced the opposite anxiety. An open coastline with no hard border felt like a permanent vulnerability. The stories gave that anxiety a shape and a villain. The genre died when the First World War made fictional invasion feel redundant. But its fingerprints are all over twentieth-century spy fiction: the sleeper agent, the compromised institution, the nation that doesn't know it's already losing.](https://newsletter.dvy.io/visit/9pLEQcCs) [Surveyors in heavy coats work in a snowy, barren landscape beside a theodolite and tripod, with an umbrella and dog nearby.](https://newsletter.dvy.io/visit/pVw4m68e) [The 19th-century obsession with measuring Earth's exact shape](https://newsletter.dvy.io/visit/pVw4m68e) Measuring the shape of the Earth sounds like it should be a single clever experiment. It turned out to be a 40-year international effort involving thousands of triangles, Arctic expeditions, and obsessive precision. The core problem: Earth isn't a perfect sphere. It bulges slightly at the equator and flattens at the poles, a shape called an oblate spheroid. By the 19th century, scientists knew this in theory. What they didn't know was how much flattening, and getting that number required measuring the actual curve of the planet's surface across thousands of kilometres. The method was triangulation. If you measure a baseline distance very precisely, then use angles to chain triangles across a landscape, you can calculate distances you can't physically walk. Swedish astronomer Nils Selander and his predecessor Friedrich Wilhelm von Struve did this along a line of longitude stretching from the Black Sea up to the Arctic, a distance of roughly 2,800 km. That arc required 258 interconnected triangles, light signals bounced between hilltops, altitude measurements taken continuously with spirit levels and barometers, and survey rods laid end-to-end hundreds of times across the ground to establish baselines. Similar efforts were running in parallel across Europe, North America, India, and Japan. It wasn't one person's eureka moment. It was a collective, multi-generational slog through logistics, weather, politics, and arithmetic. The value wasn't just the final number, it was the infrastructure of trust in measurement that the whole enterprise built. Random [Cut-away diagram of a mechanical pencil showing the inner spring, lead tube, and tip mechanism.](https://newsletter.dvy.io/visit/ZLYZhaaV) [Cut-away diagrams of everyday objects](https://newsletter.dvy.io/visit/ZLYZhaaV) [Slice anything in half and suddenly you understand how it works. This site does exactly that, with detailed cut-away illustrations showing the insides of ordinary objects, things like mechanical pencils, cameras, and door locks that most of us use without a second thought about what's happening inside. It's a nice reminder that almost every mundane object around you is a small feat of problem-solving, with springs, cams, gears, and tolerances all working together so you don't have to think about any of it.](https://newsletter.dvy.io/visit/ZLYZhaaV) [Landing page for artmyvibe with the headline “Find art to match how you feel” over a dark collage of paintings and a mood search bar with tags like joyful, melancholic, and lonely.](https://newsletter.dvy.io/visit/hYMam9Vx) [What if a painting could describe how you're feeling right now?](https://newsletter.dvy.io/visit/hYMam9Vx) A search engine for art, but instead of typing in an artist or title, you describe a mood. Melancholic? Restless? Quietly hopeful? It matches your feeling against a library of over 10,000 public-domain paintings and surfaces ones that fit. It's a surprisingly effective way to stumble across work you'd never have searched for deliberately. [Close-up of an old Latin title page with large serif text naming Galileo Galilei.](https://newsletter.dvy.io/visit/wtcEbxQT) [How a $10 million Galileo manuscript fooled the experts](https://newsletter.dvy.io/visit/wtcEbxQT) In 2005, a copy of Galileo's Sidereus Nuncius appeared on the market with an extraordinary claim attached: this wasn't just a rare 1610 first edition, but a unique copy signed and illustrated by Galileo himself. Experts authenticated it. A two-volume scholarly study followed. The price tag floated around $10 million. Then the clues started stacking up. Sidereus Nuncius ("Starry Messenger") is the 60-page book in which Galileo first described the Moon's cratered surface, four moons orbiting Jupiter, and the vast number of stars invisible to the naked eye. It was the visible proof that Copernicus was right and Earth was not the centre of everything. Around 550 copies were printed by a Venetian press in 1610; roughly 150 still exist. A signed, hand-illustrated copy would have been genuinely extraordinary. The forgery held together long enough to fool some of the field's most respected names. Part of what made it work is the market itself: rare science manuscripts have become serious assets for tech billionaires, which pulls books into private collections and away from the scholars and institutions best placed to scrutinise them. Provenance (the documented ownership history of an item) gets murky when enough money is moving fast enough. Worth a read if you've ever wondered how authentication actually fails, and what it costs when it does. Design [A young man holds up an abstract photograph featuring chaotic white, purple, and yellow light streaks on a black background.](https://newsletter.dvy.io/visit/G3BF6dWP) [Cosmic rays as a camera: photography student sends film to the edge of space](https://newsletter.dvy.io/visit/G3BF6dWP) [Tom Liggett, a photography student at Arts University Bournemouth, sends unexposed film up to around 30km (roughly three times commercial cruising altitude) in a sealed bag attached to a helium balloon. Up there, above most of the atmosphere, cosmic rays hit the film instead of light. When the balloon bursts and the film comes back down, he develops it normally and gets something genuinely strange: abstract curves, contours, and bursts of colour shaped entirely by radiation rather than a photographer's eye. The idea came from a simple question: what happens if you send film to space? Liggett had already experimented with dental and hospital X-rays to see how radiation affects unexposed film (airport scanners are a known hazard for photographers), but cosmic radiation at near-space altitude is a different beast entirely. As far as anyone knows, nobody has done this before. The images aren't composed or framed. Nobody pressed a shutter. The "camera" is the void, and the sensor is a strip of chemistry that's been floating above almost everything.](https://newsletter.dvy.io/visit/G3BF6dWP) [A poster with colourful hand-painted “LOREM IPSUM” and the caption “(it’s not latin.)”, surrounded by papers and paint pots.](https://newsletter.dvy.io/visit/KK5WUaEK) [Lorem ipsum is garbled Cicero, and nobody noticed for 500 years](https://newsletter.dvy.io/visit/KK5WUaEK) "Lorem ipsum" has appeared on design mockups, website templates, and print layouts for decades. It looks like Latin, and it is, sort of. It's a scrambled extract from De Finibus Bonorum et Malorum, a philosophical treatise Cicero wrote in 45 BC. Somewhere in the intervening centuries, a typesetter chopped it up, shuffled the words, and used it as filler text. Nobody seems to have noticed, or cared, for around 500 years. The connection to Cicero was only confirmed in 1994, when a researcher traced "lorem ipsum" back to a specific passage. Before that, generations of typesetters, designers, and printers had been casually reproducing mangled classical Latin without the faintest idea where it came from. [Web](https://newsletter.dvy.io/visit/sdxH7vrQ) · [Twitter](https://newsletter.dvy.io/visit/NQeXb4Q3) · [LinkedIn](https://newsletter.dvy.io/visit/eiRFmZqD)