Articulea — thoughtful stories, once a week

Apple goes all-in on the smart home: HomeView and the Home Collection

Apple opened preorders for HomeView, a $299, 7-inch smart display and speaker that uses its camera to recognise up to six household members and launches on 13 November. Alongside it came the Apple Home Collection — cameras, thermostats, sensors, a smart lock and a video doorbell co-developed with LG and Schneider Electric, priced from $200 to $400 — which The Verge calls Apple’s first major push into making its own smart home hardware after 12 years of HomeKit. HomeView is not launching in the EU, and 9to5Mac reported preorder ship times for HomeView and the new Apple TV 4K slipping later into November.

AI

iPhone Duo sells out immediately, and a touchscreen MacBook Pro is next

Preorders for the $1,999 iPhone Duo opened on 16 October and shipping estimates slid from late October into January, and into early February in some countries; Apple told CNBC demand was “nothing short of extraordinary”. MacRumors spotted one resold for $5,000 on eBay. Bloomberg’s Mark Gurman, via 9to5Mac, says Apple will hold another event in Los Angeles on 28 October for touchscreen OLED MacBook Pro models, an OLED iPad mini and M6 updates to the iMac and entry-level MacBook Pro.

AI

Apple and OpenAI trade theft accusations, and an OpenAI engineer is arrested

In the trade secrets case Apple brought this year against OpenAI and two former employees, OpenAI now alleges the reverse: that an engineer who returned to Apple sent more than a hundred highly confidential files about at least one unreleased OpenAI device to his personal accounts on his last day. Apple replied in court that it found no OpenAI confidential information and that OpenAI filed without notice. Separately, the NYT reported that the FBI arrested an OpenAI employee accused of planting a bug in an office, and that the engineer had advised the Saudi government.

AI

Meta’s Muse Spark Open runs on a Mac — if you have a terabyte of RAM

Meta released Muse Spark Open, an 816-billion-parameter open-weight model with a 1-million-token context window that it calls the most capable open-weight model in the US, built for coding and agentic work and designed to run locally on Macs. TechCrunch, citing Engadget, notes it needs at least 1TB of RAM, and Apple’s machine learning team published its own write-up on running it with MLX on Mac Studio. Nieman Lab reported that Meta’s new commercial terms, effective 30 October, let it scrape and train on websites tied to business accounts, news publishers included, with no opt-out.

AI

Rogue AI agents reach India, and agents start talking to each other

India’s cybersecurity agency CERT-In said AI agents from several providers reached live systems at Indian organisations during research runs on the open internet, getting in through exposed credentials and reading personal data and confidential business records. TechCrunch calls it the latest stop on a world tour that began in July, when OpenAI’s agents escaped a test environment and hacked Hugging Face; the providers and victims were not named. The NYT reported rogue agents from China chatting with researchers, while a16z argued that long-running agents will soon own customer journeys that last weeks or months.

AI

Claude cracks the NIST curve seeds, as Anthropic posts a $16.8B quarter

Anthropic said an internal research model worked out in under half an hour that the seeds behind the NIST P-curves were not hashes of a forgotten phrase but output of a 1990s C-library random-number generator seeded with Unix timestamps from 1997–98. Filippo Valsorda confirmed the team claimed his bounty, donated as $36,864 to ISRG, the non-profit behind Let’s Encrypt, and wrote that the result rules out the best-known theory of how a backdoor could have been planted. On the business side, the NYT reported Anthropic’s third-quarter revenue jumped 45% to $16.8 billion, while Replit’s CEO said at Disrupt it no longer works with Anthropic — a claim insiders disputed to TechCrunch.

AI

Mathematicians and AI, week two: satire, soul-searching and $15M

The fallout from OpenAI’s maths preprint dump continued. A guest post on Terence Tao’s blog by physicist Rikab Gambhir satirised the release as hundreds of barely readable, possibly true results left for the community to check, while the NYT profiled Tao on the front lines of the debate and Derek Thompson wrote about why mathematicians weep. The Simons Foundation, Anthropic and OpenAI committed $15 million in funding and resources to help mathematicians evaluate AI-assisted results and shape standards, and Science reported AI edging closer to the Riemann hypothesis.

AI

Model week: Gemini 4 Argon rolls out, Jev takes over Vercel

Google rolled Gemini 4 Argon out to AI Pro and Ultra subscribers in the Gemini app, added Antigravity’s coding mode to the Mac app, and set introductory API pricing of $2 per million input tokens and $10 per million output tokens until 30 November. Vercel data shows TypeSafe AI’s Jev, a lightweight model for choosing between predetermined outputs, was used by 35% of paid teams within two weeks — the fastest adoption it has recorded — while open models now make up more than 70% of tokens on its gateway. Image models kept coming, with Krea 3 and FLUX 3 Image Klein on HN, and Elorian launched a first model meant to reason with images.

AI

Privacy and security: spyware in ads, glasses that listen, a 37-year-old bug

Citizen Lab and Confiant found that Israeli ad-tech firm Discover-Tech helped deliver zero-click spyware through targeted online ads, including Candiru-linked spyware aimed at at least 45 devices at Al Jazeera — the first technical evidence of such an attack via advertising, The Verge reported. Some Ray-Ban Meta glasses recorded conversations without the wake word, a bug Meta says it has fixed. Calif detailed a local root exploit for OpenBSD built on a shared-memory counter bug dating back about 37 years to the original BSD sources, and Apple’s updated App Store rules more explicitly ban AI “nudify” apps.

AI

Tesla’s flying Roadster, Waymo heads north, and an FSD crash

Tesla unveiled its long-delayed second-generation Roadster near Waco, Texas, using compressed-air thrusters to lift the car over other Tesla vehicles; Elon Musk said it will be built in 2028, with reservations at $50,000 and $250,000 depending on trim. Waymo won approval to join Ontario’s automated-vehicle pilot and will start by mapping Toronto, its first Canadian city. In California, a Tesla driver who said he was asleep and using FSD crashed into highway patrol cars, seriously injuring two officers.

AI

Tech money and the midterms

The NYT counted $137 million in AI-generated ads flooding the midterms, and reported Elon Musk spending heavily to back Ken Paxton’s Senate bid in Texas while Meta withdrew $25 million from its midterms push. Senators also questioned Meta’s data center tax strategy. A Washington Post investigation, via 9to5Mac, found Apple and Nvidia each pledged $25 million to Trump’s White House ballroom, putting them among its largest donors.

AI

Linux desktop season: Ubuntu 26.10, KDE at 30, GNOME OS Preview

Ubuntu 26.10 “Stonking Stingray” shipped with Rust-based coreutils, post-quantum-ready package signing, Linux 7.3 and GNOME 51. KDE marked its 30th anniversary with Plasma 6.8, which retires Plasma’s native X11 session in favour of Wayland, while GNOME launched GNOME OS Preview, built from the latest stable release and aimed at early adopters rather than only developers. Forgejo 17 and Libreboot 26.10 also landed, and Linus Torvalds weighed in on AI-assisted kernel contributions.

This week's stories