TL;DR
China, the United States and the European Union are activating or completing different AI oversight systems within a 19-day period. China uses regulatory approval, the EU relies on risk-based conformity rules, and the US framework offers voluntary government evaluation tied to trusted-partner status.
China, the United States and the European Union are bringing three different AI oversight systems to key implementation points within 19 calendar days, according to a regulatory dispatch from Thorsten Meyer AI. The sequence begins with China’s rules for human-like AI services on July 15, continues with a US voluntary evaluation framework on August 1 and reaches the EU AI Act’s full application date on August 2.
China’s Interim Measures for AI Anthropomorphic Interaction Services, issued in April by five government bodies, extend the country’s existing approval model to companion AI and agent-like systems, the dispatch reported. Covered services face security reviews before public deployment, algorithm registration and continuing duties after release.
The Chinese framework gives regulators a direct role in system operation. According to the dispatch, authorities can request algorithm changes, while providers must report security incidents within 24 hours and answer government information requests within 48 hours. The exact application of those requirements across different products has not yet been detailed in the source material.
In the United States, the dispatch describes Executive Order 14409 as establishing a voluntary pathway under which participating frontier-model developers provide the government with 30 days of pre-release access. Evaluation criteria are classified, and trusted-partner status serves as a procurement incentive. The framework is not described as a legal requirement to obtain approval before releasing a model.
The EU AI Act follows a third model: risk classification, conformity reviews, technical records and post-market monitoring. Its staged application began with prohibited practices in February 2025, followed by obligations for general-purpose AI models, and is scheduled to reach full application on August 2, 2026.
Three Gates Close in Nineteen Days
The Pre-Release Regime Goes Global
Same-day-verified · one instinct, three architectures — and none of them binds the open frontier
Anthropomorphic-interaction measures take effect: five agencies extend the CAC approval regime to companion AI and agents.
EO 14409’s classified benchmark and voluntary 30-day pre-release framework harden. NSA designates covered frontier models.
The AI Act becomes fully applicable — the staged rollout that began February 2025 reaches its final station.
Same instinct, three theories of a gate
STEELMAN: THE GATE-SKEPTIC CASE
Pre-release regimes structurally favor incumbents who can afford the process — and none of the three binds an open-weight release from a lab outside its jurisdiction. The gates go up exactly as the fastest-moving part of the frontier walks around them.
The signal: a model can clear all three gates having been evaluated for three almost non-overlapping things — content control, fundamental rights, national security. Jurisdiction is now an architectural property. If your deployment calendar doesn’t carry July 15, August 1, and August 2, it’s a calendar for a market you’re not in.
The Confidence Advantage: Optimizing Privacy, Cybersecurity and AI Governance for Growth
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
One Calendar, Three Regulatory Tests
The dates reveal a shared policy judgment that some powerful AI systems should face government scrutiny before public release. They do not represent agreement on what that scrutiny should measure. China focuses on content control and social stability, the EU emphasizes fundamental rights and product safety, and the US framework centers on national-security risk.
For developers operating across these markets, one model release may trigger separate review tracks with different evidence, timelines and legal consequences. Passing one jurisdiction’s process does not establish compliance elsewhere because the three systems examine largely different categories of risk.
The dispatch also argues that the cost of formal reviews may favor large, well-funded developers. Smaller laboratories may lack the legal, testing and documentation resources needed to manage several regimes at once, although the source provides no cost estimates showing the scale of that burden.
Ai Engineering Made Practical: Build Reliable Ai Systems With Retrieval, Tools, Evaluation, Monitoring, And Safety—So Teams Ship Faster With Less Risk
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Three Systems Differ
China has required public-facing generative AI services to undergo security reviews and algorithm filing since 2023. The new measures apply that established approach to anthropomorphic interaction, including services designed to behave like companions or autonomous agents.
The EU system is broader but organized around risk categories and market conformity, rather than approval for each individual use. General-purpose models that cross the act’s systemic-risk threshold face added evaluation, reporting and incident-management duties.
The US approach remains the least binding of the three. Its central mechanism is a voluntary access window, with government procurement and trusted-partner status providing incentives. The United Kingdom, by comparison, continues to rely on sector regulators and nonbinding principles without a formal economy-wide pre-release checkpoint.
“Every serious jurisdiction has concluded that some class of AI system should meet the state before it meets the public.”
— Thorsten Meyer AI dispatch
AI-Native Platforms for Agentic Systems: A Practical Guide to Runtime Architecture, Evaluation, Governance, and Enterprise Operating Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Deadlines and Coverage Still Unsettled
The EU timetable could still change. 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-system deadlines. According to the dispatch, the package still requires Council adoption and publication in the Official Journal, so August 2 remains the operative date unless that process is completed.
Details of the US framework are also shielded from public examination because its benchmark criteria are classified. The source does not identify which models the National Security Agency will designate as covered, which developers will participate or what happens when a model performs poorly.
The reach of all three systems over open-weight models released abroad remains uncertain. The dispatch contends that none fully binds a laboratory outside its jurisdiction, but enforcement may depend on where a model is offered, deployed or integrated into commercial services.
Theory and Applications of Image Registration
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Regulators Move From Rules to Enforcement
Attention now shifts to China’s implementation of the new measures, US decisions on covered frontier models and the EU’s handling of the Digital Omnibus. Developers planning releases across the three markets will need to track registration, evaluation and conformity deadlines separately as regulators publish further guidance and begin enforcement.
Key Questions
Do the three jurisdictions require the same pre-release review?
No. China uses a binding approval and registration structure, the EU applies risk-based conformity duties, and the US system is voluntary under the framework described by the dispatch.
The cited framework does not establish a general approval requirement. It offers a 30-day government evaluation window for participating developers and uses trusted-partner status and procurement access as incentives.
Has the EU AI Act deadline already moved?
No change is treated as final in the source material. The Digital Omnibus still needs Council adoption and Official Journal publication, leaving August 2 as the operative date for now.
Which AI services are covered by China’s new measures?
The measures target anthropomorphic AI interaction services, including companion-style systems and agents designed to simulate human interaction. The dispatch does not provide a complete list of covered product categories.
Are open-weight models covered by these regimes?
Coverage remains unsettled, particularly when a model is released by a laboratory outside the regulator’s jurisdiction. Obligations may still arise when that model is marketed or deployed locally.
Source: Thorsten Meyer AI
