Token Enters the Livestream Room for the First Time: How to Keep AI's Problem-Solving Ability in Check?
This year's "Double 11," Token entered the livestream room for the first time and became a purchasable commodity, because every conversation and instruction between users and AI consumes Token. As of the first half of 2026, the number of generative artificial intelligence users in China exceeded 700 million, with a penetration rate of over 50.0%. The article proposes that AI governance should shift from "extinguishing fires after the fact" to "moving security forward," imposing rigid requirements such as filing and testing on high-risk scenarios such as healthcare and finance, and utilizing flexible tools such as regulatory sandboxes and algorithm audits.
Topics
Body · AI translation
Article body
This "Double 11," many people's order lists included a new, invisible and intangible item: Tokens.
In the past, livestreams often rang out with "3, 2, 1, link up," but this year there's a new saying: "3, 2, 1, Tokens up." Tokens entered the livestream room for the first time, giving this year's "Double 11" a new face.
Why is everyone buying Tokens? The reason is simple: every conversation we have with AI and every instruction we give AI consumes Tokens. You could say Tokens are AI's "fuel."
Paying for "fuel" shows that AI has truly taken off. The data proves it: the "Generative Artificial Intelligence Application Development Report (2026)" shows that as of the first half of 2026, China's generative AI user base exceeded 700 million people, with a penetration rate of over 50.0%.
81-year-old Turing Award winner Martin Hellman made an interesting analogy: humanity is like someone who just got a driver's license and somehow got their hands on a 500-horsepower sports car.
A 500-horsepower sports car, in the automotive world, is a high-performance level—its power is self-evident. So some people have turned their thoughts to AI for crooked purposes. A local court published a typical case—lawbreakers used AI to generate dynamic facial videos and broke through a platform's facial verification. Cases like this are not few.
So, how to keep a 500-horsepower sports car running on the right track is the key.
Just as driving in the real world has traffic rules, AI on the road also needs rules. But for AI, the difference is that AI technology iterates quickly and risk forms are ever-changing; traditional "one-size-fits-all" or after-the-fact punitive regulatory approaches easily fall into the dilemma of "crack down and it dies, let go and it's chaos." Therefore, a dynamic, layered, self-adjusting governance ecosystem needs to be built. The regulatory approach must also shift from "putting out fires after the fact" to "moving safety forward," embedding governance rules and requirements into front-end links such as design, development, and verification. Only in this way can risks be controlled from the source.
Effective regulation, in the author's view, should also combine firmness with flexibility—accelerate when it's time to accelerate, brake when it's time to brake. For high-risk scenarios such as healthcare and finance, rigid requirements such as filing and testing should be implemented, while for low-risk scenarios, more reliance can be placed on compliance self-testing and industry self-discipline. In addition, flexible tools such as regulatory sandboxes and algorithm audits can also be fully utilized, allowing innovation to trial and error within a controllable space and enabling regulators to identify risks in advance.
(Source: People's Daily Overseas Edition) Mid-article embedded advertisement