China approves mandatory autonomous-driving safety standard

China's market regulator just drew a regulatory line under two of AI's most safety-critical uses: driving cars and diagnosing patients. Beijing approved a mandatory national standard for autonomous driving system safety alongside new quality rules for AI medical devices, part of a 338-standard batch that signals how the country intends to govern AI where failures have physical consequences.

China's market regulator approved a mandatory national standard for autonomous driving system safety, alongside new quality standards for AI medical devices. The State Administration for Market Regulation said it approved 338 national standards — 15 of them mandatory — covering high-tech, manufacturing, and daily life. The headline item is the safety-requirements standard for autonomous driving systems in intelligent connected vehicles (GB 44721—2026), which sets rules for how such systems are designed, developed, and tested so they operate reliably in complex traffic and cut accidents caused by system failure or misjudgment. It takes effect on July 1, 2027, giving automakers and suppliers a runway to comply. The batch also includes a national standard on quality requirements and evaluation terminology for AI medical devices, the opening piece of what regulators describe as a national standards system for medical AI — establishing a common vocabulary for development, testing, and quality control.

The move is a deliberate sequencing choice: China is standardizing AI where it touches physical safety before deployment becomes ubiquitous, rather than after an incident forces the issue. The 2027 effective date matters too — it lands right as autonomous driving services are scaling in major Chinese cities, meaning the rule will bite exactly when the technology is most exposed. Medical AI gets a softer touch for now — terminology and quality evaluation, not hard clinical mandates — but it lays the foundation for the enforcement that typically follows.

Insta360 is pushing AI into its cameras, shipping a voice assistant powered by Alibaba's Qwen on the mainland. The Shenzhen action-camera maker announced it is rolling out "Kira," an AI voice assistant for its GO Ultra thumb camera. Voiceprint recognition and intent detection run on-device, while the cloud handles conversation, mode switching, and translation. In mainland China the assistant reportedly runs on Alibaba's Qwen model family, a pairing Chinese tech media framed as "putting Qwen in a camera" — a distribution experiment for Alibaba's model beyond chat apps and into consumer hardware. The move extends a broader AI-hardware push: Insta360 also sells the Connect, an AI dual-camera conferencing device integrated with DingTalk, and its Q1 revenue grew 83 percent year over year to 2.48 billion yuan. Camera makers are increasingly treating the AI assistant, not the lens, as the differentiator — and for Qwen, every Insta360 owner is a potential endpoint.

What to watch: whether GB 44721—2026's 2027 deadline pushes China's robotaxi operators to publish compliance roadmaps this year, and whether other consumer hardware makers follow Insta360 into Qwen-powered devices.

Mandatory safety standards for machines that drive and diagnose — do you think regulators should draw the same line for AI in finance or hiring? Tell us in the comments.

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

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

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day

Perplexity open-sources its agent sandbox after 118 failed escapes

Perplexity handed its agent containment layer to the open-source world, and Washington's AI task force stopped reading briefings and started working. Perplexity open-sourced SPACE on Thursday — the sandbox it built to keep its own AI agents from getting out — and says nine frontier models failed to break out of it across 118 attempts. Each agent runs inside a small virtual machine with its own operating system, with credentials and network access managed outside the sandbox so an agent can re

Read more

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

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

Amazon's new Bee devices listen all day — for $11.99 a month

Perplexity open-sources its agent sandbox after 118 failed escapes