On 2 June 2026 the US President signed an executive order strengthening AI model review. According to media reports, an arrangement that would have allowed the US government to obtain access to certain AI models for up to 90 days before release was, after industry lobbying, shortened to a 30-day pre-release review window — described as a “compromise” executive order.
The core change
The order adjusts the “pre-release review window”: under certain conditions the government obtains access before a model is publicly released, to assess safety risk. Cutting the window from 90 to 30 days was the result of a contest between industry and regulators.
Notably, even shortened to 30 days, the direction is unchanged — review moves from “after-the-fact accountability” to “before-the-fact intervention”. This logic shares a spirit with China’s filing, registration and security assessment for generative AI, though the tools and paths differ.
Subsequently, after safety incidents involving unauthorised use of US government systems at an AI company, the White House further announced mandatory AI safety incident reporting, requiring AI companies to report to federal authorities immediately upon a safety incident involving their models. US AI safety regulation thus shows a turn “from voluntary to mandatory”.
Why this matters to Chinese AI companies
Chinese AI companies commonly go global in two ways: products directly entering the US market, or indirect entry through APIs, open-source models or technology licensing. Both paths touch the new requirements of US AI regulation.
First, if a model or service is deployed in the US, pre-release review and safety incident reporting obligations may apply directly. This means product release schedules must reserve time for regulatory windows.
Second, if a model is distributed as open source, companies need to assess whether downstream use triggers US-side compliance duties, and the intersecting constraints of open-source licences and export controls.
Third, the shift of safety incident reporting from voluntary to mandatory means companies need internal mechanisms for incident identification, classification and reporting. Such mechanisms are emphasised on the Chinese side too — see our AI law hub for related analysis.
Going global under three parallel tracks
For Chinese AI companies, going global can no longer be handled with a single per-jurisdiction approach. The EU AI Act sets duties by risk tier, the US advances through a combination of executive orders and departmental rules, and China builds access and ongoing supervision through filing, registration, security assessment and dedicated legislation. The shared trend across all three is movement from “after the fact” to “before the fact”, with emphasis on traceability and verifiability.
For comparison, see our EU AI Act service page and our analysis of the dual-regulation overlay of the outbound-investment rules and the EU AI Act. The latter discusses exactly the “domestic review plus overseas compliance” dual-review structure, which this US turn now makes equally valid in the North American direction.
Three things companies can do now
First, build regulatory windows into product schedules. Whether the target is the US or the EU, pre-release review and compliance assessment take time, and product roadmaps should reserve buffer.
Second, establish cross-border safety incident response mechanisms. Clarify who identifies, who reports and within what time frame, rather than organising after an incident.
Third, map the model distribution chain. Open source, API and licensing carry different duties, and each should be checked against the actual distribution method. For support on going-global compliance paths, cross-border incident response and multi-jurisdiction arrangements, see our Multi-Jurisdiction Compliance and EU AI Act services.
This article is based on the June 2026 US AI model review executive order and public reporting. It is for reference only and does not constitute legal advice. The specific application of foreign law should be based on local legal advice.