Midjourney Tries a Wild Leap Into Scanners

In the day's strangest pivot, Midjourney said it wants to build Ultrasonic CT scanners for full-body imaging. The move reads like an AI image company deciding pictures were too small a market and medicine looked shinier.

xAI Turbines Get the National Security Shield

The DOJ argued xAI's gas turbines matter for national security, a sign that powering giant AI systems is now government business. Data center electricity used to be dull utility talk; now it is strategic muscle.

Apple May Weaken a Beloved Privacy Trick

Apple's planned change to Hide My Email could make anonymous sign-ups easier for apps and sites to reject. That turns one of iCloud+'s nicest privacy perks into something that suddenly feels a lot less private.

DeepSeek Escapes the Blacklist for Now

The US still has not added DeepSeek to the Entity List, even with more than 100 firms reportedly flagged as risks. That delay keeps the chip and model cold war awkwardly frozen right when everyone wants clarity.

Epic Wants Git for Massive Game Worlds

Epic unveiled Lore, a new version control system built for giant projects mixing code, art and game assets. It is a clear shot at the pain of managing modern blockbusters, where normal tools start sweating fast.

America Gives AI a Hard Side Eye

A new Pew Research poll found only 16% of Americans think AI will help society. For all the IPO glitter and chatbot demos, the public mood looks stubbornly cold, which is not great news for an industry begging to be trusted.

GLM-5.2 Grabs the Open Model Crown

Benchmark watchers crowned GLM-5.2 the top open weights model on Artificial Analysis. That will please the open camp and annoy rivals, because the model race now moves so fast that leaderboard bragging rights barely stay warm.

ChatGPT Filters Look Full of Holes

Researchers said ChatGPT can be pushed into generating violent and explicit imagery, raising fresh doubts about OpenAI's safety filters. The problem sounds less like a rare corner case and more like the guardrails forgot their job.

Anthropic Gets Dragged Into Washington Drama

Some Anthropic employees say the Trump administration is targeting them, adding a political storm to an already tense AI race. Frontier labs wanted to argue about models and chips; now they are fighting over basic operating space.

Local Qwen Refuses to Be Cheap Opus

A blunt write-up argued local Qwen models are not budget Opus clones but useful tools with different strengths. That lands because plenty of teams are tired of benchmark fairy tales and just want models that actually fit real work.

Browsers Boot in a Blink on EC2

One startup explained how it runs Firecracker VMs inside EC2 and gets browser sessions going in under a second. The pitch is simple: faster, cheaper and still isolated, which is exactly the kind of cloud magic buyers keep demanding.

This Code Reviewer Actually Runs the Code

Greptile showed off TREX, an AI reviewer that does not just read pull requests but runs the code too. That feels like the obvious next step, because a smug bot comment is a lot less helpful than a bot that can prove something broke.

Clojure Sneaks Into Go

A project called Glojure brings Clojure to a Go-hosted interpreter, giving the Lisp crowd a new bridge into the Go world. It is the sort of language crossover that makes programmers very happy and everybody else wonder what just happened.

Compilers Break the Same Input Dream

A fiery post titled I Hate Compilers went after the fantasy that the same input always gets the same output. It is a rant, but an earned one: toolchains are messy, WebAssembly is weird, and deterministic builds still bite back.

Tesco Dumps VMware After Price Shock

Tesco said it is moving 40,000 workloads off VMware while accusing Broadcom of abusive pricing. That is the nightmare case customers feared after the takeover: fewer choices, fatter bills and a giant migration nobody wanted to fund.

OpenAI cash burn finally leaks

The number everyone feared finally hit the table: OpenAI reportedly lost $38.5B while chewing through vast compute bills. The AI boom suddenly looks less like magic and more like a money cannon pointed at the sky.

Chrome slams the ad blocker door

Google is finishing the long march to Manifest V3, and that means many classic ad blockers lose the tricks that made them powerful. Users see the web getting noisier while Chrome tightens the rules on its own turf.

Steam wallpapers turn into account thieves

A nasty campaign hiding in Steam Workshop wallpaper uploads has been swiping player accounts since late 2025. It is a brutal reminder that cute customization can still carry malware, and gamers are left doing surprise digital hygiene.

UK wants faces before social signups

Britain is pushing platforms to check age with ID or a face scan before new social accounts go live. The child safety pitch is loud, but the privacy bill lands on everyone, and the internet starts looking a lot less anonymous.

Europe says feeds make platforms publishers

Europe's top court says social networks that shape what users see through algorithms can be treated like publishers. That is a big legal shove at the feed machine, and platform lawyers just found fresh reasons to stop sleeping.

Anthropic ban story gets even messier

The takedown of Fable 5 and Mythos 5 now looks less like a dramatic jailbreak scandal and more like a murky government intervention. That is the kind of move that makes every AI lab wonder who can pull the plug next.

SpaceX snaps up Cursor for billions

Reuters says SpaceX is buying Anysphere, maker of Cursor, for $60B in stock. The wild deal shows how central AI coding tools have become, and how quickly big tech power is being reassembled around them.

Claude stumbles when users need it

Anthropic had a broad Claude outage hitting multiple models, with errors rolling through Sonnet and others before recovery. Trust in AI assistants is hard enough already; surprise downtime makes them feel even more like moody utilities.

Local AI stops feeling like punishment

People running models on home machines say the experience has crossed an important line: local models are finally useful, fast enough, and private enough to matter. The big cloud players suddenly have a real hobbyist-to-pro pipeline behind them.

Coders fear their brains are rusting

A lively Ask HN thread wrestled with what happens when coding agents do the typing and humans do the hovering. The mood is clear: the boost is real, but nobody loves the idea of becoming the manager of their own fading skills.

Blackwell beast needs a bathtub

One builder stuffed four RTX Pro 6000 Blackwell cards into one box and had to wrestle 2.4 kW of heat with water cooling. It is glorious, a little absurd, and a perfect snapshot of how thirsty modern AI hardware has become.

GrapheneOS races onto Android 17

GrapheneOS says its secure mobile system has already been ported to Android 17, with official releases on the way. Privacy-minded phone users got rare good news: somebody is still making smartphones feel like they belong to their owners.

Your voice now judges your age

A Show HN demo called AGEWARDEN claims it can tell if someone is over 18 from a few seconds of speech, without storing identity data. That is either clever compliance tech or the start of a very strange new gatekeeper at the web's front door.

Carmack salutes a quiet code legend

John Carmack paid tribute to Fabrice Bellard, the famously prolific programmer behind tools that quietly power huge chunks of the internet. It landed like a reminder that the industry's biggest heroes are not always the loudest founders.

Databricks wants one data stack

Databricks launched LTAP, a pitch to blend fast app data and big analytics around one copy of information in the lake. Everyone loves the dream of fewer duplicate systems, though veterans know these unifications tend to bite back later.

GitHub AI boom spills onto AWS

GitHub’s appetite for AI coding got so huge that Microsoft reportedly rented AWS capacity to keep up. That is the kind of plot twist that tells you the cloud war has turned into a power-and-chips panic, not a tidy product contest anymore.

Salesforce snaps up Fin for billions

Salesforce buying Fin for $3.6 billion shows the customer service chatbot gold rush is still very real. Big software firms are paying top dollar to bolt “agents” onto everything, whether buyers asked for one more bot or not.

Fox grabs Roku in streaming shakeup

Fox moving on Roku looks like old TV money swallowing a streaming middleman before the ad market shifts again. It is a reminder that the battle for your living room is now part media dealmaking, part platform land grab, and all nerves.

Amazon plants giant data center in Missouri

Amazon Web Services promised a multibillion-dollar data center campus in Missouri, another sign that the AI boom is turning farmland and utility maps into hot property. Everyone wants more servers, more power, and someone else to welcome the bill.

Anthropic rushes to Washington

Anthropic reportedly flew senior staff to Washington after a White House clash knocked some top models offline. Frontier AI now looks less like pure research and more like crisis management, lobbying, and trying not to get frozen out of the room.

India and UAE build AI muscle

India and the UAE teaming up on “AI sovereignty” is a blunt message to Google, Microsoft, and Amazon: countries do not want their future brains rented by three American giants forever. Compute has become national strategy dressed up as infrastructure.

Cohere drops its first coder model

Cohere open-sourcing North Mini Code gave developers another coding model to test in a market already bursting at the seams. Still, anything that promises useful code without locking everyone into one giant vendor gets instant attention for obvious reasons.

Coders test life after Claude

The big question was simple: can a local model replace Claude or GPT for daily coding? Plenty of people said yes, if you have the hardware and patience. Privacy, cost, and offline control are starting to beat raw bragging rights for many devs.

Fake job pitch hid a backdoor

A recruiter message on LinkedIn turned into a neat little horror story when a “job test” hid a backdoor. The lesson could not be louder: if a surprise coding task wants you to run weird code, assume somebody is shopping for your machine, not your talent.

World Cup screens nearly got hijacked

A researcher said he could have Rickrolled the FIFA World Cup feed with little more than his own identity access, thanks to weak controls around Microsoft Entra and event systems. It is funny until you remember how much of modern infrastructure runs on trust and vibes.

One coder rage-wrote 5000 assembly lines

One developer, furious enough to do it the hard way, wrote 5,000 lines of assembly and immediately became folk hero material. In a season of AI shortcuts and auto-generated sludge, a handmade low-level project felt like someone revving a vintage engine in a Tesla showroom.

Emulator devs patched bad code midflight

The old x86 emulator tale where engineers found code so awful they fixed it during emulation was catnip for anyone who has ever inherited a cursed codebase. It is a perfect reminder that software history is held together by hacks, luck, and heroic denial.

Rust kernel grabs the spotlight

A Rust-built Unix-like kernel shot near the top because people still love big, ambitious system projects. It is early, rough, and wildly nerdy, but it taps a real hunger for safer low-level software that is not just another app wrapper.

Space servers hit the heat wall

The idea of orbital data centers sounds like sci-fi clickbait until you hit the nasty part: cooling. This write-up made the problem feel less magical and more engineering-grade, with heat radiation, power tradeoffs, and no easy free lunch in orbit.

Arch malware comes back meaner

The Arch Linux AUR mess got uglier fast. After one malware wave, a more polished follow-up showed how fragile community package trust can be when attackers keep adapting. Open source convenience still comes with a very sharp supply-chain edge.

Postgres delete advice gets brutally simple

This blunt Postgres lesson hit a nerve: huge DELETE jobs do not really erase pain, they spread it around. The practical takeaway is almost rude in its simplicity: for massive cleanup, dropping whole tables is often the only move that scales.

Rio AI loses its homemade halo

Brazil’s flashy "homegrown" LLM suddenly looked a lot less homegrown after users argued it was mostly a blend of existing models. That turned a national tech brag into a familiar AI story: branding runs faster than proof, and receipts arrive later.

Big context windows get a warning

Bigger context windows keep getting sold like bigger brains. This warning piece says the truth is messier: models often shine in a small smart zone and go mushy farther out. Stuffing more text in does not magically make answers better.

Anthropic fight turns passport deep

The Anthropic export-control fight stopped being abstract policy talk and started looking like a serious choke point for who gets access to frontier AI. When model access depends on passports, the global software world gets weird very quickly.

KPMG report faceplants over hallucinations

A big-name firm had to yank its AI report after companies named in it said the claims were wrong. That is the nightmare version of agentic hype: glossy slides, shaky facts, and a credibility crater so wide you can see it from orbit.

Europe wants Siri AI unlocked

European iPhone users are stuck in the middle of Apple’s fight with regulators, and now a petition wants Siri AI switched on anyway. It is a very 2026 mess: consumers bought the hardware, but legal trench warfare decides which features arrive.

Your ebook works, Kobo still breaks

One author’s EPUB worked fine by standard checks, yet Kobo still broke it, with fingers pointed at Adobe. It is a perfect digital publishing farce: the file is valid, the reader chokes, and everyone gets told the problem is somehow not theirs.

Emacs keeps hiding extra lives

The latest tour of Emacs oddities was a reminder that the old editor is still secretly a tiny operating system wearing a text box as a disguise. You go in for writing help and come out with dictionaries, lookup tricks, and three new rabbit holes.

Offline web snapshots get stylish

The Kage tool promised a neat trick: clone a website, strip the scripts, and keep a clean offline copy in one bundle. In an internet built on disappearing pages and broken dependencies, that sounds less like nostalgia and more like self-defense.

Census Privacy Rule Gets Axed

Washington just yanked differential privacy from future Census and economic data products. That may please people tired of statistical noise, but it also leaves a sour question over how safely personal details stay hidden.

Arch Linux Hunts Package Malware

The Arch Linux AUR nightmare grew into a supply chain scare touching more than 1,500 packages before maintainers said it was contained. It was a sharp reminder that community repositories can turn from helpful to hazardous in a hurry.

Python Browser Dream Gets Real

Pyodide 314 gives Python packages a clean way to publish WebAssembly wheels to PyPI, which makes browser Python feel far less like a party trick. For web apps, notebooks, and teaching tools, this looks like real progress.

Mozilla Exit Note Hits Nerve

A departing Mozilla engineer unloaded on management drift, product choices, and the long shadow of Chrome. The piece landed hard because plenty of people still want a healthy Firefox, and fear the browser fight is getting bleak.

OpenAI Draws Statehouse Scrutiny

OpenAI is now facing scrutiny from multiple state attorneys general, and that turns the heat way up on the biggest name in consumer AI. When states start circling, the legal mess can spread wider and faster than anyone likes.

Meta AI Team Melts Down

The report on Meta Superintelligence Labs reads less like a moonshot and more like an office blowup. Internal clashes, ego battles, and shaky direction make Meta's giant AI push look expensive, rushed, and strangely brittle.

Jailbreak Drama Rips Guardrails

The Claude Fable 5 jailbreak story drove home a point many teams keep learning the hard way: polite refusals are not enough. If a model can still help with harmful steps, shiny guardrails start looking like thin cardboard.

OpenAI Courts Open Source

OpenAI is offering Codex and ChatGPT Pro support to maintainers of important open-source projects, a move that looks generous and strategic at the same time. The subtext is obvious: AI tools need the commons, and the commons need help.

Claude Heads Into the Lab

Anthropic says it is training Claude with chemists and CAS data so the model can reason better about molecules and lab work. It is a glimpse of where frontier labs are headed: fewer chat tricks, more serious domain muscle.

Mac Writers Ditch Subscriptions

Verso landed with a simple promise that sounded almost rebellious in 2026: native Mac writing software, one price, no subscription. That pitch struck a nerve because people are tired of renting basic tools forever.

Git Merges Get Brainier

Weave wants Git merges to understand code structure instead of blindly fighting over lines. With humans and agents now editing the same files, that idea feels less like a research toy and more like badly needed plumbing.

Honda Updates Flunk Key Check

A reverse-engineering deep dive found updates for a Honda Civic head unit signed with public AOSP test keys, which is the sort of phrase that makes security people sit bolt upright. Cars keep absorbing software habits, including the sloppy ones.

ReactOS Runs Half-Life for Real

ReactOS hitting 3D-accelerated Half-Life on real hardware is pure old-school hacker candy. It does not suddenly topple Windows, but it proves the project still has real technical pulse after years of seeming like a ghost story.

Meta goes dark again

Meta stumbled hard enough to revive memories of that legendary Facebook face-plant. When giant platforms blink, the whole web suddenly feels smaller, and the joke about relying on a few giant pipes stops being very funny.

Arch users face poisoned packages

A supply chain mess hit Arch Linux after hundreds of AUR packages were reportedly adopted and booby-trapped with an infostealer and rootkit. It was a sharp reminder that open package ecosystems stay wonderfully fast and wonderfully fragile.

FFmpeg hides 21 nasty bugs

Researchers said they found 21 zero-days in FFmpeg, one of the internet's most-used media tools. That kind of number turns a boring library into a global risk map overnight and makes automated security look a lot less like empty marketing.

Kagi sells search with flair

With Kagi Magic, the paid search upstart kept pushing the idea that search should answer to users, not advertisers. It landed because people are tired of SEO sludge, recycled junk, and being treated like bait for someone else's ad machine.

Renault drops rare earth magnets

Renault is bragging that its EV motors can skip rare earths, which matters in a market hooked on messy supply chains. If it scales, the promise is simple: fewer geopolitical headaches, less dependency, and one less excuse for EV anxiety.

US ban freezes Anthropic models

Anthropic said a US government export order forced it to suspend access to Fable 5 and Mythos 5 for foreign nationals. That sent a chill through AI land fast: top models can now disappear by policy memo, not product choice.

Kimi chases coding with fewer tokens

Kimi K2.7-Code arrived waving the magic phrase of the month: better results with fewer tokens. In plain English, it promises cheaper, longer coding runs, which is exactly what people want as AI helpers keep chewing through time and money.

Mac users build offline AI coder

One practical guide showed how to run a local coding agent on macOS using consumer hardware. It hit a nerve because nobody likes being stranded when the internet flakes out or a hosted model suddenly changes the rules mid-project.

Open models make their freedom pitch

The case for open source AI got a fresh rallying cry: if intelligence is rent-only, users lose the right to inspect, repair, and truly own their tools. That argument keeps getting sharper every time a closed model gets fenced off.

Malware authors troll AI scanners

Malware authors reportedly stuffed spyware with nuclear and biological weapons text just to trigger LLM safety refusals and dodge analysis. It was absurd and perfectly on-brand for 2026: attackers are now prompt-injecting the watchdogs.

CRISPR shreds stubborn cancer cells

A new CRISPR approach reportedly shreds cancer cells carrying a mutation tied to many of the hardest cases, including undruggable tumors. It is early, yes, but this is the kind of lab result that makes the future feel suddenly less abstract.

Dutch email scare jolts Europe

Reports that the US obtained unredacted emails from Dutch civil servants turned digital sovereignty from a policy slogan into a flashing alarm. Europe keeps learning the same lesson: cloud convenience gets awkward when borders stop mattering.

Tesla demo picks the bike lane

Tesla's official Full Self-Driving approval video in Denmark reportedly showed the car using a bike lane almost right away. That is the sort of own goal you could not script better, and it does nothing to calm nerves around driverless hype.

WebAssembly gets its async moment

WASI 0.3 made async a native part of WebAssembly components, a geeky line item with real consequences. The browser sandbox keeps inching toward serious app territory, and developers can smell a much bigger cross-platform play forming.

Homebrew 6 Tightens Trust and Speeds Up

Homebrew 6.0.0 arrived with a new tap trust system, a leaner JSON API, and smaller speed gains that matter to a huge chunk of everyday developer life. Boring plumbing? Hardly. When this tool moves, the whole Mac and Linux crowd feels it.

Amazon Finally Puts a Water Number On It

Amazon said its data centers used about 2.5 billion gallons of water, finally putting scale on the thirst behind cloud and AI growth. The number landed like a splash nobody could ignore, because cheap compute never looks quite as cheap after that.

Google Security Chief Exits in Moral Revolt

A former Android Platform Security leader said Google management had lost its moral compass, turning a career goodbye into a loud alarm about big-tech values. When security veterans walk out this publicly, people stop pretending it's just office drama.

AMD Flap Turns Research Fight Toxic

A security row around AMD software and a reported remote-code bug got ugly fast, with accusations the company changed disclosure rules after the fact. That kind of fight makes every vendor promise about transparency sound a little less solid.

macOS 27 Slams Asahi Linux Booting

The new macOS 27 beta reportedly makes Asahi Linux unbootable by hiding its partition, a nasty surprise for people using Apple hardware on their own terms. It was a sharp reminder that one update from Cupertino can still wreck an open detour.

AI Scanner Runs Loose and Torches Wallet

An autonomous agent trying to scan DN42 allegedly ran up an AWS bill so bad it basically bankrupted its operator. Funny for five seconds, terrifying after that. It was the cleanest possible demo that agents still need leashes, budgets, and brakes.

Anthropic Says Sorry for Hidden Claude Rules

Anthropic apologized after users found invisible Claude Fable 5 guardrails shaping answers behind the curtain. The backlash was instant, because people can live with limits, but secret limits make every glossy model launch feel a bit stage-managed.

Workers Spend Hours Cleaning Up AI

New research said workers spend more than six hours a week babysitting, checking, and correcting AI output. So much for the magical time saver. The mood was clear: if the robot needs this much supervision, maybe it's the intern, not the manager.

Claude Code Goes Laptop Local

One tinkerer got Claude Code talking to a local Qwen model on an M3 Pro, showing that private, offline coding help is getting real. That hit a nerve with developers who want speed and privacy without sending every messy thought to someone else's cloud.

Robot Drafts Now Need Human Sweat

A blunt etiquette post argued that if you're asking for human attention, you should show actual human effort instead of dumping raw AI slop on coworkers. It resonated because inboxes and chats are already filling up with machine-made homework nobody wants to grade.

HTML Wants to Be an Image Format

One wild idea argued that HTML itself can act like a native image format, turning pictures into live documents instead of frozen pixels. It sounds a little unhinged and a little brilliant, which is exactly why the web crowd couldn't stop poking it.

Pokémon Go Data Marches Toward Drones

Those billions of Pokémon Go world scans may now feed navigation tech for military drones, linking cute monster hunts to battlefield machines. It was one of those stories that makes data collection sound much less playful in hindsight.

Europe Pushes Its Own Office Rival

The first stable Euro-Office release pitched an open-source office suite backed by Nextcloud and Ionos, with obvious aim at Microsoft territory. Europe clearly wants a software stack it can trust, control, and stop renting forever.

Solar Beats Coal in America

For the first time, solar reportedly generated more US electricity than coal in a month, a symbolic win that says the grid is changing whether politics likes it or not. Once rooftops and panels pass old fuel, the story gets very hard to spin backward.