China just wrote the first rulebook for AI that reads brains
While Washington spends its fourth day arguing about who gets to referee frontier labs, Beijing kept doing what it does: publishing specs. Two landed this morning — a world-first standard for brain-computer-interface devices, and a pair of AI systems pointed at mineral exploration.
China's drug-and-device regulator approved the world's first medical-device standard written for AI that processes brain signals. The National Medical Products Administration issued it Monday — the country's third brain-computer-interface device standard, and the first anywhere that governs a BCI product whose core function is running AI over EEG data. It sets technical requirements and test methods across the whole data path: collection, processing, annotation, storage and access. The stated target is a complaint device makers keep raising — that dataset quality control has no shared yardstick and build processes aren't comparable between vendors. NMPA's own framing is the part worth noting: a BCI device has to "learn" from large volumes of brain recordings before it can infer a patient's intent, so data governance is a clinical safety property, not administrative paperwork. The standard takes effect September 1 next year, which gives the sector roughly a year to get its training corpora auditable. Whoever writes a category's first standard usually ends up owning its tolerance ranges, and this lands about six months after Beijing approved the world's first commercial invasive BCI implant. It's also the same pattern we tracked at daybreak — China just told AI customer service it cannot promise you a refund: binding rules first, arguments later.
China switched on two homegrown AI systems for finding ore, and the claim on the table is that five days of machine work replaces more than a month in the field. The pair, reported by 21st Century Business Herald, splits the job the way an agent stack would: an "intelligent mapping" system that updates an AI geological map progressively as survey crews collect data in the field, flagging geological bodies the eye misses and planning routes so crews walk less ground at lower risk, then merging everything into a final compiled map for review. The second is a dedicated prospecting system that predicts where deposits are likely to sit, including a kilometre-deep targets that outcrop at the surface only faintly, if at all. The strategic logic is domestic: mine surveying is manpower-heavy, expensive, and tied to critical-mineral security, which is why the pitch is framed as exportable — a template other countries' geological surveys could adopt. The honest caveat is that accuracy numbers aren't published, and a geological map is exactly the kind of artifact where a confident hallucination becomes a dry well three years later. Worth watching anyway, because this is applied AI measured in avoided fieldwork rather than benchmark points.
What to watch: whether any government drafts the boring-enforceable alternative the labs' critics keep asking for — incident reporting, exploit-specific tests, and deployment-based thresholds — instead of a compute line only incumbents can clear.
Should the first international standard for brain-reading devices come from a national regulator, or from a body with other countries at the table? Tell us in the comments.
