China orders labels and a human sign-off on every AI broadcast
China spent this month putting AI-generated television in front of viewers. Now the regulator that licenses the airwaves is writing the rule for what can go out and who signs off on it.
China's broadcast regulator will require identifying labels on any programming made with AI, and a human in the loop at every key production step. At a State Council Information Office briefing on Thursday, Liu Jianguo of the National Radio and Television Administration said programs generated, produced or aired with artificial intelligence "must carry a corresponding label," that key stages of production and the review of the finished cut "must insist on manual review and approval," and that AI remixing of existing footage — the practice Chinese outlets call AI 魔改 — is banned outright. The framing was deliberately two-sided: encouragement for innovation "has a scale," and regulated development "has a boundary," with the rules to be tightened as the technology moves. It is the first time NRTA has put the label, the human gate and the anti-remix ban in one policy breath.
The label is the easy half. China's content labeling rules already require marks on AI-generated material, and the all-AI drama that took a prime-time slot this month — we covered it in September — China's prime-time TV slot now belongs to an all-AI drama — is exactly the kind of production this is aimed at. The harder half is the accountability clause: "key production stages and the review of the finished content must insist on human review." That turns a technology question into a paperwork question. A broadcaster now has to be able to show who approved which cut, which quietly rules out the obvious cost saver — a pipeline where a model drafts, edits and packages a segment and nobody senior watches it before air. Read alongside the labeling mandate for short video platforms already in force, NRTA has drawn a line that leaves AI everywhere in the workflow except the final sign-off, and a production sector that has spent two years measuring AI by how much of the human it removes now has to budget the human back in.
Brain-interface companies have already raised more than $1 billion in 2026, beating the previous four years combined. PitchBook's figure, reported by the Financial Times, puts this year's total past $1 billion against $1.56 billion across the four years before it, and the narrower pictures agree on the shape. Neurofounders' 17-company snapshot of companies that actually record neural signals found $1.75 billion raised in the last 12 months, against just under $1.6 billion in all prior years for the same firms; Inside BCI counted $960 million across nine rounds in the first quarter alone. The clearest signal in the mix is who is writing the cheques: OpenAI took the largest position in Sam Altman-backed Merge Labs's $252 million seed at an $850 million valuation, Science Corp raised a $230 million Series C at $1.5 billion, and Chinese state-linked funds are moving into the domestic cohort. Valuations are concentrating hard at the top — Neuralink alone is estimated near $9 billion. Funding, not revenue, is the story: essentially none of these companies sells anything yet.
What to watch: whether NRTA defines what counts as a "key production stage" and whether it attaches penalties, and whether Science Corp's PRIMA retinal implant clears its European CE mark decision to become the first brain-interface product on a commercial market.
If a broadcaster must show who approved the final cut, does the same accountability logic apply to an AI agent that ships code or answers a customer? Tell us in the comments.
