Apple's Siri AI home hub is in employee testing, eyeing October

Share
Apple's Siri AI home hub is in employee testing, eyeing October

Apple's first AI-era hardware is quietly living in its own employees' houses, and Zhipu is selling Chinese enterprises a privacy switch with a few visible holes in it.

Apple is testing its long-delayed home hub in employees' homes and aiming to ship it between October and early 2027, according to Mark Gurman's Power On newsletter. The device, known internally as J490, is a roughly square screen that either sits on a stand — Gurman describes it as looking like a HomePod mini cut in half — or mounts to a wall. It runs a new operating system built around Siri AI, pairs with an iPhone for setup, and borrows interface pieces from iOS, tvOS and watchOS: its own home screen, widgets, clock faces. A front-facing camera identifies who is looking at it and personalises what appears — apps, notes, calendar entries, news, music — and Gurman says Apple manages that recognition without a Face ID-grade depth sensor.

Siri is the whole pitch. Apple's reported demos have the assistant searching across a user's devices and accounts, finding a specific photo, resuming a song or podcast a friend sent, and carrying out precise tasks in Notes, Reminders and Calendar. That framing surfaced in John Ternus's first keynote as CEO, where he described the iPhone as an "intelligent personal hub" for the AI era — Gurman reports the presentation was convincing enough that people inside Apple thought he was about to unveil the hub itself. The same newsletter says Apple has started cutting staff at Fitness+ as it reassesses that service.

The hardware has reportedly been finished for months; the delay has been the assistant. Apple shipped iOS 27 this month with its rebuilt Siri AI still labelled a beta, and a device that exists to be talked to inherits every one of those shortcomings at launch. An Echo Show-style screen with a smaller display is not a novel product — what Apple is actually testing in those homes is whether Siri can hold a conversation well enough that anyone wants a second always-listening Apple device in the kitchen. If it can, the hub becomes the beachhead for the home security and robotics plans already circulating for 2027. If it can't, Apple will have shipped a very nice clock.


Zhipu announced that its MaaS platform will add a no-retention mode for enterprise and developer traffic, days after its coding agent was caught uploading whole workspaces. The mechanism is opt-in through the MaaS console: once active, the platform says it will not statically store user inputs and outputs, and data is used only to complete the current call. Zhipu is framing it as the strictest such policy among Chinese model providers — a direct answer to the ZCode episode, where a developer showed the agent packing entire repositories and Git history into an archive only Zhipu could open, prompting an apology in the company's user group.

The exclusions are where this gets interesting. Batch API and File API are explicitly out of scope, because those products need platform-side persistence to work at all, and Zhipu says it may retain data for 30 days or longer where regulations require it or where it is investigating suspected abuse. That is a normal carve-out and a real one: a compliance team reading the announcement will notice that the workloads most likely to contain sensitive bulk data — the batch jobs and uploaded files — are precisely the ones the switch does not cover. Zhipu also promised to open-source ZCode, and did — though only the post-fix version, which we covered this morning in Zhipu open-sources ZCode — but only the version after the fix. Retention is now a feature Chinese labs compete on, and the fine print is the product.

What to watch: whether Zhipu extends the option to Batch and File APIs, and whether other Chinese labs match it before buyers start writing it into procurement terms.

Would a no-retention switch change which model API you build on — or does it need to cover batch traffic too? Tell us in the comments.

Read more

Houmo moves compute inside memory — and A20 enters production

Houmo moves compute inside memory — and A20 enters production

Beijing chipmaker Houmo used its partner conference this week to lay out two compute-in-memory routes at once — and to say the first chip built on the newer one is already in production. Houmo's A20, its first chip built on a 3D compute-in-memory architecture, has entered production. The part supports up to 16 chips fully interconnected, claims a 50% improvement in compute utilization over the previous generation, and covers models from 3B to 30B parameters, according to co-founder and product

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model