TL;DR
China, the United States and the European Union are activating separate AI pre-release regimes within a 19-day period. The rules share an interest in examining certain systems before public use, but differ sharply on approval, enforcement and policy goals.
China, the United States and the European Union are activating three different forms of pre-release AI oversight within 19 days, forcing developers serving those markets to account for separate government approval, evaluation and conformity processes. The sequence began in China on July 15 and continues with US and EU milestones on August 1 and August 2.
China’s Interim Measures for AI Anthropomorphic Interaction Services took effect on July 15 after being issued in April by five government agencies, including the Cyberspace Administration of China. According to the source dispatch, the measures extend China’s existing approval structure to companion AI, human-like services and agents, building on security assessments and algorithm registration already required for public generative AI services.
The dispatch says China’s framework allows regulators to request algorithm changes before deployment and imposes continuing duties after a service launches. These reportedly include a 24-hour deadline for reporting security incidents and a 48-hour response period for government information requests. The new measures retain Beijing’s use-by-use regulatory approach, under which the state can influence system design and operation.
In the United States, an August 1 framework linked to Executive Order 14409 is described as offering developers a voluntary 30-day government evaluation window based partly on classified benchmarks. The National Security Agency is expected to designate covered frontier models, while trusted-partner status would act as a procurement incentive. One day later, the EU AI Act reaches full applicability, bringing its risk classification, documentation, conformity assessment and post-market monitoring structure into its final scheduled phase.
Global Launch Calendars Split
The close timing creates a practical problem for companies releasing the same model in several markets: one launch may face three separate reviews that examine different risks. China’s regime focuses on content controls and social stability; the EU framework centers on fundamental rights and product safety; the US system described in the dispatch focuses on national security.
The result is not a common global safety test. A model could pass all three processes after being examined against largely different criteria. Developers may need jurisdiction-specific model configurations, records and release schedules, making regulatory geography part of system architecture rather than a final legal check before launch.
As an affiliate, we earn on qualifying purchases.
Three Models of State Review
China has required security assessments for public generative AI services since 2023, according to the dispatch. Its model is an approval gate: covered providers complete an algorithm-registration process, may be directed to alter a system and receive a filing number for a public registry. The July rules carry that structure into anthropomorphic AI interactions.
The EU uses a broader conformity-before-market system rather than approving each individual use. Its AI Act has entered force in stages, beginning with prohibited practices in February 2025 and general-purpose AI obligations in August 2025. The United States has adopted the lightest of the three structures described here: a voluntary access window backed by classified testing and government procurement incentives, rather than mandatory permission to release a model.
“One instinct, three architectures.”
— Thorsten Meyer AI dispatch
AI safety and security assessment software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Deadlines and Coverage Remain Fluid
The US framework’s practical reach remains uncertain because participation is described as voluntary, its central benchmarks are classified and the covered frontier models depend on NSA designations. Public information does not yet show how many developers will participate, how trusted-partner status will affect procurement or whether Congress will make the 30-day process mandatory.
The EU schedule also carries a live qualification. The source says a Digital Omnibus package, approved by the European Parliament on June 16 by 423 votes to 57 with 174 abstentions, would move some high-risk deadlines. It has not yet completed Council adoption and Official Journal publication, so August 2 remains the operative date under the source’s account. The three regimes also leave potential gaps for open-weight models released by organizations outside their jurisdictions.
As an affiliate, we earn on qualifying purchases.
August Deadlines Test Enforcement
Attention now turns to the August 1 US milestone, including NSA coverage decisions and signs of developer participation. On August 2, companies subject to the EU AI Act must work from the deadlines still in force unless the Digital Omnibus completes the remaining legislative steps. Regulators’ first enforcement actions, filing decisions and public guidance will show how these three pre-release systems operate beyond their legal texts.
generative AI model evaluation kits
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What changed on July 15?
China’s anthropomorphic-interaction measures took effect, extending its regulatory structure to covered human-like AI services, companion systems and agents.
Does the United States require approval before an AI model launches?
Not under the framework described in the source. It provides a voluntary 30-day evaluation window, with classified benchmarks and procurement-linked trusted-partner status serving as incentives.
Has the EU’s August 2 deadline been postponed?
No postponement is legally effective under the supplied account. Proposed changes still require Council adoption and Official Journal publication, leaving August 2 as the current operative date.
Do the three regimes test the same risks?
No. Their priorities overlap only partly: China emphasizes content and social stability, the EU focuses on rights and product safety, and the US process targets national-security concerns.
Are open-weight models covered by all three gates?
Not necessarily. The dispatch identifies a possible enforcement gap for open-weight releases originating outside a regulator’s jurisdiction, and the reach of each regime will depend on its coverage rules and enforcement practice.
Source: Thorsten Meyer AI
