On 10 April 2026 the Safety Governance Committee of the Artificial Intelligence Industry Alliance (AIIA) held its first 2026 meeting in Wuhan, under the theme “collaborative governance, safeguarding the frontier”. Participants included Peking University and Huazhong University of Science and Technology, alongside model and security players such as Tencent, Baidu, Huawei, Zhipu, MiniMax, iFlytek and StepFun.
Why industry-side self-regulation matters
Rule-making for AI governance has long been led by regulators. But since 2025 a clear trend has emerged: industry-side self-regulation and coordination mechanisms are forming quickly — standards working groups, safety governance committees and open-source ecosystem consensus platforms are appearing in rapid succession.
These mechanisms do not create binding obligations directly, but they shape the industry’s consensus on what compliance means. When leading vendors and universities discuss risk tiering, evaluation methods and emergency response at the same table, that discussion settles into group standards and practice guides, which are then cited by regulators and written into customer procurement requirements. For smaller companies, industry consensus is often an earlier and more concrete signal than regulation.
The focus of safety governance
From public information, the committee’s agenda centres on model security, content security, risk tiering and evaluation, and emergency response. This aligns with TC260’s establishment of the AI Safety Standards Working Group (WG9), which is advancing standards such as a maturity-assessment method for AI safety capability.
The division of labour is roughly this: the standards working group sets standards, the safety governance committee promotes implementation. For companies, the implication is that evaluating model safety capability will increasingly rely on quantifiable methods rather than self-description.
What it means for compliance work
First, make safety capability verifiable. Participating in evaluations, keeping test records and maintaining a risk-tiering ledger will be more persuasive than an assertion that “our model is safe” — see our AI law practice.
Second, watch early versions of standards and consensus. Group standards and practice guides often precede the formal national standard, and aligning early avoids detours when the details land.
Third, trace supply-chain security upstream. The security posture of foundation models, inference frameworks and labelling providers affects the final product’s compliance, so procurement should include verification steps — see our data-compliance service.
Fourth, bring safety governance into the document system. Whether it is minutes of industry-organisation meetings or internal risk-assessment reports, these are material that can be shown to regulators and partners.
Connection to the subsequent regulatory tempo
Industry coordination mechanisms usually give direction before detailed rules. At the end of April 2026, the CAC launched the “Qinglang — Rectifying AI Application Disorder” special action, listing synthetic-content labelling and AI data poisoning as prominent problems — turning risk points discussed in the industry into enforcement priorities — see our analysis of the Qinglang AI application special action.
For AI companies, the value of industry governance platforms is not merely “attending meetings”. They offer a position from which to see the risk list earlier, and a channel to express views before rules take shape. Whether and how to participate is itself part of compliance strategy.
This article is based on public information concerning the AIIA Safety Governance Committee’s meeting of 10 April 2026. It is for general reference only and does not constitute legal advice.