US-China Model Safety Divergence Creates a Compliance Planning Gap
What happened
The Brookings Institution published How China and the US will regulate AI models for safety, a detailed examination of emerging model safety regimes in both countries. The analysis maps how each government is building requirements around pre-deployment testing for biological, chemical, nuclear, and critical-infrastructure risks, as well as ongoing evaluations after a model is released. China's approach builds on existing rules such as China's Interim Measures for the Management of Generative Artificial Intelligence Services and the China Draft AI Law. The US trajectory draws on voluntary and emerging frameworks including the NIST AI Risk Management Framework (AI RMF 1.0) and Playbook. The report finds that neither regime is converging toward a shared standard, leaving multinational organizations to navigate two separate testing and documentation tracks. This follows earlier analysis on how middle-power AI hedging creates a multi-jurisdiction compliance gap and how a US-China AI incident notification proposal creates cross-border reporting challenges.
Why it matters
- ·Organizations deploying advanced models in both the US and China must now plan for two separate safety testing tracks. Neither regime recognizes the other's evaluations as sufficient, so a single pre-deployment assessment is unlikely to satisfy both.
- ·The focus on biological, chemical, nuclear, and critical-infrastructure risk thresholds means safety testing is no longer a purely technical exercise. Legal, compliance, and risk teams must be involved in scoping what risks to assess, how to document findings, and when to escalate.
- ·Voluntary frameworks like the NIST AI Risk Management Framework (AI RMF 1.0) and Playbook may not provide sufficient evidence of compliance under China's binding rules. Relying on a single framework to satisfy both markets is a governance gap that regulators in either jurisdiction could exploit.
Stay ahead of stories like this
Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.
Recent issues
- State-linked attackers and criminal operators are actively weaponizing AI tools as entry points into enterprise systems, turning internal platforms and open-source frameworks into live attack vectors.5 Nov
- Three separate incidents this week each created a new compliance action item: verify your AI vendors' data sourcing, demand agent monitoring coverage guarantees, and audit trusted app identities in your cloud environment.29 Oct
Free every Thursday. Unsubscribe anytime.
Governance controls affected
What to do now
- ☐Map every advanced AI model your organization deploys or plans to deploy against the specific risk categories both regimes require testing for: biological, chemical, nuclear, and critical-infrastructure threats. Identify which models fall within scope.
- ☐Review whether your current pre-deployment safety testing documentation is structured to satisfy both US and Chinese requirements separately, or only one. Flag any models where a single evaluation record is being used for both markets.
- ☐Ask your legal and compliance team to confirm whether your organization's China-facing AI deployments are subject to the Interim Measures for Generative AI or the draft AI Law, and what testing records those rules require.
- ☐Confirm that your vendor due diligence process for AI model providers asks explicitly about how the vendor conducts safety testing for biological, chemical, nuclear, and infrastructure risks, and whether those tests are documented separately by jurisdiction.
- ☐Schedule a review with your risk team to assess whether your multi-jurisdiction AI compliance mapping is updated to reflect the divergence between US and Chinese safety testing expectations, and assign an owner to monitor each regime for updates.
What to watch next
Compliance teams should monitor the progression of the China Draft AI Law, which would codify safety testing obligations that currently exist in piecemeal form across sector-specific rules. In the US, the NIST AI Risk Management Framework (AI RMF 1.0) and Playbook is expected to be updated with more specific guidance on high-stakes risk categories. This could narrow the gap between voluntary and binding expectations. The earlier US-China AI incident notification proposal signals that cross-border reporting obligations may follow testing obligations. Now is the right time to build dual-track documentation infrastructure before it becomes mandatory.