A Second Round of Dialogue, Not Legislation
On August 13, 2026, market regulators in Beijing and Shanghai convened a second "platform economy breaking involution" dialogue with couriers, academics and lawmakers, and came away with concrete commitments from Meituan, Alibaba's Taobao Flash Purchase (formerly Ele.me) and JD's food-delivery unit. The platforms agreed to stop counting time riders spend stopped at red lights against delivery deadlines, cap the assumed average speed used to calculate customer arrival estimates at 15 km/h, extend delivery windows for bad weather and merchant prep delays, and — in Meituan's case — launch "Tuanbao," an AI assistant that flags accident-prone intersections, red-light countdowns and illegal turns in real time (Sixth Tone; Yicai). Pilots are already live in parts of Beijing, Suzhou, Wuxi and Qingdao, with Meituan targeting coverage of more than a million riders by year-end.
This builds on the Cyberspace Administration of China's trial "Negative List for Life Service Platform Algorithms," issued in January 2026 and formally rolled out at a CAC deployment meeting on February 3 that targeted algorithmic "black boxes, discrimination and collusion" across food delivery, ride-hailing, freight, e-commerce, travel and ticketing (CAC deployment meeting). By May, the CAC reported 63 optimization measures across thirteen platforms — Meituan, Taobao, JD, Didi, Amap, T3, Baidu, Manbang, Lalamove, Pinduoduo, Douyin, Ctrip and Qunar — including more than 700,000 Meituan smart helmets, 240,000 Taobao Flash safety devices, Didi cutting its commission cap from 29% to 27%, and Amap moving to 100% manual review of major driver-facing decisions (CAC results report).
The Case Regulators Can Fairly Make
China's delivery riders operate under algorithmic dispatch systems that, for years, treated delivery time as close to the only optimized variable — a dynamic widely documented since the 2020 exposé on riders "trapped in the system." When an app's ETA assumes a rider can average well above 15 km/h through city traffic, the gap between the algorithm's assumption and physical reality gets closed by riders running red lights. That is a genuine negative externality: the cost of an optimization choice made by a platform is borne by riders' safety and by pedestrians and other traffic. A regulator that goes after the specific algorithmic parameter — the speed assumption, the red-light penalty, the timeout fine — is arguably better targeted than a blunt minimum-wage or maximum-hours mandate, and it can move faster than legislation. Credit where due: a rider no longer has to choose between running a red light and eating a penalty is a real, measurable safety win, and CAC's own numbers (63 measures, commission cuts, hundreds of thousands of safety devices) suggest this isn't just messaging.
Where the Model Runs Thin
The mechanism producing these fixes, though, is administrative guidance and public dialogue, not statute. Nothing in the negative-list framework creates a right a rider can invoke directly if a platform quietly reverts to aggressive ETA assumptions once regulatory attention moves on — enforcement runs entirely through CAC and SAMR discretion, and CAC's own May report concedes that "some platforms still exhibit selective rectification," changing algorithms only when competitors are forced to match. That is a structural admission that compliance is closer to a repeated coordination game between the state and three dominant firms than a durable legal floor.
It also matters who is, and isn't, in the room. The August 13 dialogue included couriers, academics and lawmakers — but China has no independent rider unions with bargaining or strike leverage; the state convenes and curates who speaks for labor. Compare this to the EU's 2024 Platform Work Directive, which creates a rebuttable legal presumption of employment status and a worker's enforceable right to an explanation of automated decisions, adjudicable in court independent of any regulator's ongoing goodwill. China's approach delivers a comparable-looking fix faster, but on a foundation that can be renegotiated or quietly abandoned the next time a subsidy war resumes.
There's also an incentive problem the parameter-tweaking doesn't touch: riders are still paid per delivery under a piece-rate structure that rewards speed and volume, inside a market that SAMR itself flagged as a top-ten "involutionary competition" case in January 2026 and that its parallel "Ten Provisions" rules (published June 17, targeting prolonged subsidy wars under the Anti-Monopoly and Anti-Unfair Competition Laws) are still trying to cool (SAMR Ten Provisions). Fix the ETA math without touching the pay structure that makes speed the rider's only lever for income, and the pressure that produced red-light-running in the first place doesn't disappear — it just waits for the next price war to resurface.
What Would Make This Stick
If Beijing wants these fixes to outlast the current dialogue cycle, the negative list's binding provisions need to move from administrative guidance into enforceable rules with a grievance channel a rider can invoke without waiting for the next summit — the same durability gap that separates China's fast-moving soft law from the EU's slower but judicially enforceable model. Proportionate regulation of delivery algorithms is a legitimate government function; the question is whether it survives the next subsidy war, or needs reinventing every time one breaks out.