Quad9 DNS exits France & eMudhra quits TLS certificates - Hacker News (Oct 23, 2026)

Someone has gotten Apple's iOS, unmodified, to boot on a budget Android phone. Lock screen, home screen, Safari, all running on hardware Apple never touched. More on that in a moment. Welcome to The Automated Daily, hacker news edition. The podcast created by generative AI. I'm TrendTeller, and today is October 23rd, 2026. We've got a DNS provider leaving an entire country, AI writing mathematical proofs about real software, chatbots sharing an oddly unanimous view of humanity, and Xbox putting a clock on cloud gaming. Let's get into it.

Quad9 DNS exits France

We'll start with internet plumbing. Quad9, the nonprofit that runs a free DNS resolver which blocks malicious domains, says it will stop serving users in France on November 24th. The decision follows a French court ruling in a copyright case brought by broadcaster beIN, which would require Quad9 to block certain domains for French users. Quad9 argues that blocking them worldwide would push French law onto everyone else, and that geo-blocking is unreliable and caused trouble in earlier litigation. It also says the order would barely dent piracy while weakening security for French users who rely on the service. An appeal is pending, but for now Quad9 says pulling out is the only way to stay true to its neutrality and privacy principles. It's a reminder that DNS resolvers are increasingly becoming the front line in copyright enforcement, and some operators would rather leave a country than play gatekeeper.

eMudhra quits TLS certificates

Staying with internet trust, an update on the eMudhra certificate story. The company's board has decided to exit the global SSL/TLS certificate business entirely. eMudhra says it revoked the affected certificate authorities within 14 hours, that no end-user certificates were issued from them, and that no data was breached. It also says the TLS business made up only about one percent of revenue, squeezed by rising compliance costs and free competitors. Existing certificates will run until they expire or are replaced, replacements will be free through partners, and its digital signature and identity businesses carry on as usual.

Claude scales formal software verification

Now to AI and software correctness. Anthropic, working with a company called Theorem, says Claude produced machine-checked proofs for security properties across 123 open-source libraries. The trick was converting existing fuzz tests into formal specifications, then proving the compiled code can't crash on inputs under a set size, given stated assumptions about the environment. Along the way, Claude found and helped patch real bugs that had survived years of testing, including memory-safety issues. Formal verification has long been seen as too expensive for most projects. If language models can do the heavy lifting, security could shift from 'we tested a lot' toward 'we can prove it.' That's a big claim, and it's still early, but the bug finds make it worth watching.

NVIDIA OpenShell sandboxes AI agents

In a related vein, NVIDIA has published OpenShell, an open-source runtime for running autonomous AI agents inside a controlled environment. The idea is to let agents read files, install packages, and call APIs without giving them free run of your secrets or network. It combines kernel-level sandboxing with formal checks on policy changes before they're approved. As agents get more access to real systems, tools that put firm boundaries around them are going to matter a lot.

AI models judge humanity

On a lighter note, ModelCompare asked twelve AI models, three times each, for their most controversial opinion about humans. Twenty-nine of thirty-five answers landed on the same theme: people care more about belonging, loyalty, and status than about truth. Variations included humans rationalizing after the fact and preferring to be wrong together rather than right alone. A few outliers argued humans are more decent than they think, or value meaning over happiness, and Grok declined to answer. Image models mostly drew the same joke about phone-addicted, self-destructive humans. The takeaway: the 'controversial' view is really a very familiar critique, echoed straight back from the human writing these models learned from.

WorldTUI simulates a terminal

And one more AI experiment, this one genuinely fun. WorldTUI is a small, 360-million-parameter model trained to behave like a terminal, with no real shell or filesystem behind it. You type, and it predicts how the screen should change. The author trained it on recorded terminal sessions, then taught it to repair its own mistakes, which helped it remember files across longer sessions. It beats prompted baseline models on simple commands but stumbles on more complex combinations. It even runs in a browser. It's a neat look at where learned, interactive interfaces might go, and where they still fall short.

HackiOS runs iOS on Android

Now, that hook. HackiOS is a research project that boots Apple's open-source XNU kernel on a POCO X5 Pro, an Android phone, and runs unmodified iOS 16.5 on top. It also works in the QEMU emulator. The lock screen, home screen, gestures, Safari, Photos, Wi-Fi, sound, and camera preview reportedly work, with some GPU support. Cellular and Bluetooth don't, and sessions are temporary. The author is clear this is a research bench, not a daily-driver ROM. Still, seeing iOS on non-Apple hardware at all is a remarkable piece of low-level engineering.

Building a container runtime yourself

Sticking with low-level curiosity, a developer wrote about building their own container runtime and what it taught them. The core lesson: containers aren't little machines, they're ordinary processes wrapped in Linux isolation features, and images are essentially folders mounted as a new root. The post also explains why Docker's runc has to work around how Go handles threads, and why so much of Docker needs root. The bigger point is a good one: rebuilding a tool you use every day is one of the best ways to really understand it.

Xbox Cloud Playtime hourly pricing

Finally, gaming. Microsoft is testing Xbox Cloud Playtime with Insiders, letting players buy blocks of cloud streaming time for games they already own, roughly 24 to 40 cents an hour. Ars Technica notes that, unlike old arcade pay-per-play, you've already paid for the game itself. There's also a free, ad-supported tier with longer waits and lower quality. Stadia and PlayStation Now both drifted away from per-use pricing, so the open question is whether enough people stream often enough to make an hourly fee feel worth it.

That's the rundown for October 23rd, 2026. Links to all of today's stories can be found in the episode notes. Thanks for listening to The Automated Daily, hacker news edition. I'm TrendTeller, see you tomorrow.

Quad9 DNS exits France & eMudhra quits TLS certificates - Hacker News (Oct 23, 2026)

Our Sponsors

Today's Hacker News Topics