China's Cyberspace Administration (CAC) closed a second round of public comment on August 2, 2026 on a revised draft of the Internet Information Services Management Measures — the foundational regulation for online platforms, first issued in 2000 and now getting its most substantial rewrite in a generation. The draft, opened for comment July 3, adds a dedicated chapter on "smart information services" that would require algorithmic recommendation providers to build human-intervention and opt-out mechanisms, ban forced algorithm use and algorithmic manipulation of trending topics, and mandate transparency and impact assessments for algorithms that set gig workers' order allocation, pay, and schedules (CAC notice).
What the draft actually requires
The operative language, per the CAC's own notice and an accompanying expert interpretation, bars platforms from compelling users to accept personalized recommendations and requires "options that are not based on personal characteristics" alongside a "convenient means to refuse" algorithmic curation altogether. It separately prohibits using recommendation systems to "interfere with information presentation, manipulate rankings, or control trending topics." For platform labor, the draft requires that algorithms governing "order allocation, remuneration, working hours, and similar matters" be "transparent, fair, and reasonable," with advance disclosure and — per the CAC's interpretive note — consultation with affected workers before major rule changes take effect (CAC expert interpretation).
The draft is catching up to a campaign already underway
What makes this draft notable is that it isn't writing on a blank slate. The CAC issued an "Algorithm Negative List (Trial)" on January 1, 2026, and spent the following months pushing platforms in delivery, ride-hailing, freight, e-commerce, travel and ticketing to self-audit. By May 8, fourteen major platforms — Meituan, Taobao Flash Sale, JD.com, Didi, Amap, T3 Mobility, Baidu, Manbang, Huolala, Pinduoduo, Douyin, Ctrip and Qunar among them — had implemented 63 algorithm "optimization" measures, including at least four unconditional order-rejection opportunities per day for delivery riders (Sina Finance, May 10, 2026). The draft's gig-worker provisions largely formalize what regulators had already extracted through pressure campaigns rather than statute — a sequencing that matters for how the rule should be read: as ratification of a settled administrative practice, not an experiment.
The stakes for labor are real. China's gig economy now covers more than 200 million workers — over a quarter of the national workforce — and Beijing has set 2027 as the target for standardizing labor practices across the platform economy, with algorithm transparency as one plank of a broader plan covering wages, social insurance and extreme-weather protections (SCMP, gig worker plan).
The case for the rule, stated fairly
The strongest argument for codifying this isn't paternalism — it's that opaque, unreviewable dispatch algorithms genuinely can set working conditions no human ever formally approved: a rider's per-order pay, route, and rest windows determined entirely by a system no one outside the platform can audit, adjusted in real time with no notice. Voluntary "self-rectification" campaigns are also inherently reversible; once regulatory attention moves on, so can the fixes. A codified opt-out-and-disclosure floor gives workers and users something to point to that doesn't depend on political priority in a given quarter. The trending-topic-manipulation ban, similarly, targets a genuine problem — coordinated inauthentic amplification is a real vector, not a hypothetical one, everywhere platforms operate.
Where it goes wrong
The problem is what sits next to the legitimate provisions. "Transparent, fair, and reasonable" is not a standard — it's a discretion grant, and the CAC's own interpretive materials describe the draft's broader goal as expanding the agency's "regulatory toolkit," not narrowing it (CAC interpretation). A ban on algorithmically "controlling trending topics" sits inside the same chapter as content-moderation and account-management rules the CAC already enforces asymmetrically — against platforms and creators that surface criticism of the state, not against state-aligned amplification. Folding that prohibition into an ostensibly consumer-protection and labor chapter makes it harder, not easier, for outside observers to tell where legitimate anti-manipulation enforcement ends and speech control begins.
Mandatory algorithmic impact assessments, meanwhile, are the kind of requirement that reads reasonably on paper and becomes a de facto pre-clearance regime in practice — especially for smaller platforms and foreign entrants without in-house compliance teams built for CAC filings. The EU's Digital Services Act and Platform Work Directive show the same tension: transparency mandates that are genuinely useful when narrowly scoped tend to metastasize into compliance overhead that entrenches incumbents who can absorb the cost.
The fix isn't to abandon algorithm accountability — the underlying harms are genuine and the gig-worker provisions in particular deserve to survive in something close to their current form. It's to decouple labor-algorithm transparency, which is measurable and enforceable against a concrete standard, from public-opinion provisions that hand a content regulator new discretionary levers under a labor-protection label. As comments close and the CAC moves toward a final text, that's the seam worth watching.