On 8 February 2024 the US National Institute of Standards and Technology announced the AI Safety Institute Consortium (AISIC). Members are drawn from companies, academic institutions and non-profit organisations. The consortium works on the AI risk management framework and on evaluation and measurement science for generative AI, and connects to implementation of the Biden administration’s AI executive order.
It is not a regulator. It is a collaboration platform that puts standards, testing and industry practice at the same table. For Chinese AI companies, the significance is that the technical threshold for the US market is shifting from “complies with the text of the law” to “can be measured and compared”.
The middle layer between standards and measurement
AISIC does not issue licences or enforce. It brings members together to develop test methods, evaluation benchmarks and best practice, and feeds the results into standards and policy. Joining or not joining does not change legal obligations, but it affects who takes part in defining what “acceptable” means technically.
The value of this middle layer is converting abstract requirements into executable tests. Whether model output is reliable, or whether safeguards work, cannot be left to supplier self-description when no common test method exists.
Relationship with the executive order
The Biden administration’s AI executive order allocates risk management, safety evaluation and transparency duties across agencies. NIST develops standards and tools; AISIC supplies industry input and test scenarios. The legal effect of the order, and its later changes, depend on the US political process. Technical consensus formed through the consortium usually has a longer life.
Companies should therefore distinguish two kinds of document: one that changes with administrations, and one that settles into industry method. The first defines current obligations; the second shapes procurement and evaluation habits for years.
What it means for Chinese exporters
Companies offering models, APIs or AI products to the US market increasingly face due diligence questions such as “do you follow the NIST AI RMF” or “have you run a generative AI evaluation”. Answering them requires internal documentation, test records and supply chain information, not a statement of principle.
Supply chain due diligence follows the same pattern. The provenance of models, data, compute and components must be explainable. Projects involving sensitive sectors or government customers face higher transparency expectations. Marketing language about being “safe” rarely survives a technical review.
Contract terms are changing as well. US customers begin to write evaluation results, incident notification deadlines and model update notices into procurement agreements. A supplier unable to produce verifiable records is often excluded at negotiation, before delivery reveals a breach.
Participation and real thresholds
Participation requires time, staff and disclosure. For most Chinese companies, a more realistic approach is to track published outputs, turn evaluation methods into internal test processes and prepare matching materials for customer conversations.
Joining the consortium and conforming to its framework are different things. The first is a collaboration status; the second is evidence of technical capability. Not joining does not prevent a product from entering the US market, but failing to explain risk management will not pass customer due diligence.
Where several jurisdictions apply, assess whether the EU or other markets impose additional requirements before building separate systems for one market. The same model documentation and test records, properly adjusted, can often be reused across customer reviews.
For a comparison of regulatory paths, see our multi-jurisdiction service; for cross-border data transfer arrangements, see cross-border data.
This article is based on the announcement of the AI Safety Institute Consortium by the US National Institute of Standards and Technology on 8 February 2024, and on public information. It is for reference only and does not constitute legal advice.