What the draft does
On 8 October 2026, seven central government departments, led by the Ministry of Human Resources and Social Security (MOHRSS), released a draft on protecting the working rights of people in new forms of employment. Yicai Global describes it as a 54-article draft. The Ministries of Transport and of Culture and Tourism are among the drafters, and public comment ran to 8 November. The draft covers ride-hailing drivers, food-delivery riders and livestreamers.
The algorithm provisions are the notable part. According to Yicai, platforms would have to explain the basic principles and operating mechanisms of their algorithms to workers, and algorithmic discrimination would be prohibited. People's Daily Online, carrying China Daily's report, says the draft sets requirements to prevent discrimination in the allocation of online orders and payments. Sixth Tone reports that decisions with significant consequences, including suspending work assignments or banning accounts, must undergo human review. The same report says companies must investigate and respond within 48 hours when workers challenge pay, working hours, performance reviews or account bans.
I could not retrieve the draft text from MOHRSS's site, so this analysis relies on press accounts of its contents. The final wording may differ.
The strongest case for the draft
The case for these rules is strong. The 2024 MOHRSS guidelines noted that China has 84 million workers in new employment forms. For many of them, an automated system decides whether they work today. A platform can cut off a rider's income without notice, and the rider has no human to appeal to. Guidance that only asks platforms to "solicit opinions before changing algorithms", which is the 2024 language, is a weak check on that power. A rule that gives workers a named human, a deadline and a right to object would address a real power imbalance. A pro-innovation view should accept this. Platform labour markets need worker trust to function.
How it fits the existing regime
China already regulates algorithms, but from a different angle. The Cyberspace Administration of China's 2022 recommendation provisions took effect on 1 March 2022. In the DigiChina translation, Article 24 requires providers with "public opinion properties or social mobilization capabilities" to file details with a government system within 10 working days of starting service. Article 17 requires an explanation where an algorithm has a major influence on users' rights. Article 20 already addresses workers: providers offering work-dispatch services must protect workers' lawful rights, including remuneration and rest, and "establish and perfect" the algorithms that govern sign-on, allocation, pay and working time.
The registry's public face is thin. Carnegie's analysis of the first batch of 30 filings, released in August 2022, found a regime that emphasises security assessment more than disclosure to affected people. In practice the filing system gives regulators visibility, not workers.
The new draft shifts the audience for accountability. The 2022 rules mostly ran from platform to regulator, with Article 20 as a general instruction on labour. The draft runs from platform to worker, with a deadline attached. That is a more enforceable design, because a worker can tell whether a response arrived in 48 hours.
Where proportionality matters
The risks lie in implementation, and a good rule can still fail there.
- "Human review" can become a rubber stamp. If a reviewer approves hundreds of automated flags an hour, the rule is satisfied on paper while nothing changes. The final text should require a reviewer with authority and information to reverse the decision, and should require platforms to record outcomes. That is better than a bare duty to have a human "involved".
- "Explain the principles" needs a standard. Disclosing basic principles is workable. Disclosing ranking weights invites gaming by workers and rivals, and it can expose fraud-detection logic. A tiered duty is better: plain-language principles for workers, with fuller technical detail for regulators under confidentiality.
- Discrimination needs a definition. A prohibition on algorithmic discrimination in order and pay allocation is sensible. Dynamic pricing and demand-based dispatch are legitimate, though, and a vague ban could chill efficiency improvements that raise total earnings.
- The 48-hour clock needs triage. Fast responses for account suspensions make sense. Applying the same deadline to every pay query could swamp small platforms and favour incumbents with large support teams.
The gap between the 2022 filing regime and the 2026 worker-facing duties is an opportunity. Regulators could reuse existing filing data to check whether a platform's stated allocation principles match its filed algorithm. That would add enforcement without new bureaucracy.
What to watch
Three things will show whether this is substantive. First, whether the final text defines "major" decisions precisely enough to be tested. Second, whether the human-review duty names who reviews and what they can overturn. Third, whether enforcement ties to the CAC's filing system or sits only with labour inspectors. Keeping a worker's right to a fast answer, while leaving platforms room to keep adjusting their dispatch systems, would give China one of the more practical models of algorithmic accountability for gig work.
Other jurisdictions debating platform-work rules can learn from this draft, whatever they think of China's wider approach to online governance. The better ideas are specific ones: a deadline, a named human and a right to object. Those are easier to enforce than broad transparency mandates.