China's cybersecurity standards committee named a robotics company as lead drafting unit for one of five artificial intelligence security standards now recruiting participants

What these notices are, and what they are not

The notices are dated 3, 4, 5, 6 and 11 August 2026, a span of nine days on the committee's own posting dates. Each recruits 参编单位, drafting-participant units, for an artificial intelligence security national standard, and all five sit under the same working group, WG9.

A participant call is upstream of everything a compliance reader normally cares about. There is no draft text, no threshold, no penalty and no commencement date, because the standard has not been written. What a call does tell you is where a standards body has decided to spend its next year of effort, and who it trusts to hold the pen.

That second question turns out to be the interesting one here.

Who leads what

The embodied-AI security guide names Unitree, sealed on the notice as 宇树科技股份有限公司, as lead unit, with applications returned to the company itself. Unitree is a Hangzhou robotics company that builds quadruped and humanoid robots, which makes it a commercial manufacturer rather than a laboratory or a university. The on-device large-model cybersecurity guide is led by Pujiang National Laboratory in Shanghai. The AI model development security guide is led by Beijing Zhongguancun Laboratory.

The security requirements for AI training and inference frameworks name China University of Political Science and Law as lead unit. That is a law university convening the drafting of a technical software standard, and it is the pairing most likely to surprise a reader who assumes these documents are convened by engineering bodies alone.

The open-source model security guide is led by Zhejiang University in Hangzhou. Its window closed on 14 August 2026.

Taken together the five map the areas China is standardising next: robots that act in the world, models that run on the device rather than the cloud, the frameworks models are trained and served on, how models are developed, and how they are opened.

The five deadlines, and why they may not be an opportunity

These are five different dates and should never be summarised as one. On the dates printed on the notices: the open-source model window at Zhejiang University closed on 14 August 2026; the embodied-AI window at Unitree and the on-device model window at Pujiang National Laboratory close on 21 August 2026; the model-development window at Beijing Zhongguancun Laboratory and the training-framework window at China University of Political Science and Law close on 31 August 2026.

Read those dates alongside the eligibility rule in the next section, because on their own they invite the wrong inference. All five notices require the applicant to be an existing WG9 member unit, so these dates describe the timetable of an internal committee process rather than an opening for organisations at large. A reader who is not already a member unit has nothing to act on here, whatever the date says.

One qualification on the word closed. It means only that the date printed on the notice has passed. We did not contact any lead unit and did not confirm that recruitment actually ended, or that a window was not extended.

Each notice asks for a sealed application form returned to the lead unit, and each requires the participating organisation to assign one fixed technical specialist to take part in the drafting throughout, rather than rotating staff through meetings.

What the notices say about openness, and what condition 1 requires

Each of the five notices opens with the same stated purpose, in the committee's own words: 为切实做好网络安全国家标准编制工作，鼓励更多单位切实参加到标准编制过程中，提高标准编制工作的开放性、公正性、透明性，提升标准的实用性和质量. In English, to do the work of drafting national cybersecurity standards properly, to encourage more units to take part in the drafting process, and to improve the openness, fairness and transparency of that work as well as the practicality and quality of the standards.

Each notice also describes itself as a 公开征集, which translates as a public call. Set against that, the first participation condition in all five is 该标准所属（WG9）工作组的成员单位: the applicant must already be a member unit of the WG9 working group.

Both of those come from the same document. We report the juxtaposition rather than characterise it. We have not established how many organisations are WG9 member units, and we make no claim about whether this eligibility rule is usual or unusual for standards bodies, in China or elsewhere. What it does mean concretely is that an organisation reading these notices cold cannot simply apply, and any description of them as open to industry at large would be inaccurate.

The remaining conditions are ordinary: accept the committee's participation rules, complete the tasks the lead unit assigns, attend drafting meetings on time and contribute, and assign fixed technical staff.

Why a foreign reader should care about a participant call

National standards in this family eventually become the reference points that Chinese regulators and procurement rely on, and in some cases become mandatory. Watching them at the drafting stage is the only point at which the shape of a standard is still open, and the only point at which the choice of lead unit is visible.

For a robotics or on-device model vendor the practical takeaway is that the technical direction of these five standards will be set over the coming months by the named lead units, and that their published outputs are the documents to track.

There is nothing to comply with today. There is something to watch.

What we did not verify

We did not read any draft standard. TC260 did not publish draft text with these five notices, and we did not establish whether any draft exists internally. Every description above is of the recruitment notice, not of a standard's content.

We did not verify the internal composition of WG9, so we cannot say how many organisations are eligible to apply, or whether any foreign-invested entity is a member unit.

We did not contact any lead unit, and we do not know how many applications each received or whether any window was extended. A closed window in this article means the date printed on the notice has passed, not that we confirmed the recruitment ended.

We did not verify what the five standards will be numbered, whether they will be mandatory or recommended, or when any of them will issue. None of that is stated in a participant call.

The notice bodies are published as images rather than text. We read the images directly and took the lead units, deadlines, contacts and seal dates from them.

Related briefings

• China's earlier sector AI-security drafting call
• The GB standard on AI security classification and grading
• TC260's AI agent deployment security guide
• China's rules for human-like AI companions
• AI Regulation News hub