TL;DR
China, the United States and the European Union are bringing distinct AI review frameworks into effect within 19 days. The systems share an interest in reviewing powerful AI before public deployment, but differ sharply on approval, enforcement and regulatory goals.
China’s rules for human-like AI services took effect July 15, opening a 19-day period in which the United States and European Union are also scheduled to activate major AI review milestones. The three systems all place government scrutiny closer to deployment, but they test for different risks and impose sharply different obligations on developers.
China’s Interim Measures for AI Anthropomorphic Interaction Services extend the country’s existing approval structure to systems such as AI companions and autonomous agents, according to a regulatory timeline published by Thorsten Meyer AI. The April rules were issued by five bodies, including the Cyberspace Administration of China, the national development and industry ministries, public security authorities and the market regulator.
The dispatch says covered services can face security reviews before public deployment, algorithm registration and regulator-requested design changes. It also describes continuing duties, including a 24-hour incident-reporting window, responses to government information requests within 48 hours and implementation of ordered algorithm adjustments.
In the United States, the source says Executive Order 14409 establishes an August 1 milestone for a classified benchmark and a voluntary 30-day government evaluation window. Participating frontier-model developers would provide access before release, while trusted-partner status and procurement treatment would serve as incentives. The European Union’s next milestone follows on August 2, when most remaining provisions of the EU AI Act are scheduled to apply.
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 Developer's Playbook for Large Language Model Security: Building Secure AI Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Three Gates Test Different Risks
The closely grouped dates do not amount to a shared global AI standard. China focuses heavily on content controls, public security and social stability. The EU uses risk classification, conformity checks and technical documentation tied to market access. The US framework described by the source is voluntary and centered on national-security evaluation.
For developers operating across these markets, a single model could face three largely separate review processes. Passing one would not establish compliance in another because the systems examine different conduct, documentation and harms. That makes jurisdiction part of product design, affecting release schedules, model access, monitoring and incident-response plans.
The burden may fall unevenly. Large developers are more likely to have the staff, infrastructure and legal resources needed for multiple pre-release processes. Smaller companies could face higher barriers, while open-weight models released beyond a regulator’s reach may avoid parts of all three systems. The source argues that this gap could leave the fastest-moving segment of AI development subject to the least direct review.
The Developer's Playbook for Large Language Model Security: Building Secure AI Applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI Oversight Moves Before Deployment
China has required security assessments and algorithm filings for some public generative-AI services since 2023. The July measures apply that approach more directly to anthropomorphic systems, where services simulate personalities, emotional relationships or human-like interaction.
The EU AI Act has followed a staged schedule. Its first prohibitions began applying in February 2025, followed by obligations for general-purpose AI models in August 2025. The August 2026 milestone brings most of the remaining framework into operation, including duties covering high-risk systems, market monitoring and enforcement.
Washington has taken a lighter approach in the framework described by Thorsten Meyer AI. Instead of requiring regulatory clearance, it offers participating developers a government evaluation period tied to classified criteria. The United Kingdom remains outside this pattern, retaining a principles-based system led by sector regulators rather than a formal cross-economy release gate.
“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 Prompts for Safety Professionals: Save Hours on Risk Assessments, Incident Reports, Toolbox Talks, and Safety Documentation Using Artificial Intelligence
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Deadlines and Coverage Remain Fluid
The EU schedule could still change for some systems. The source says a Digital Omnibus proposal, approved by the European Parliament on June 16 by 423 votes to 57, with 174 abstentions, would move certain high-risk deadlines. It has not taken legal effect without Council adoption and publication in the Official Journal, leaving August 2 as the operative date described by the source.
The practical reach of the US framework also remains uncertain because participation is voluntary and its core benchmark is classified. Public information does not show how many developers will participate, how trusted-partner status will be awarded or what happens when a model performs poorly. It is also unclear how any of the three jurisdictions will handle foreign open-weight releases that originate outside their direct authority.
Mercury Alert AI Senior Fall Monitor | 24/7 Passive Monitoring | Hands-Free Emergency and Safety Detection Device | Subscription Required
24/7 AI PASSIVE MONITORING: Detects falls, wandering, and nighttime movement without wearables, buttons, or check-ins.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
US and EU Milestones Approach
Attention now shifts to the August 1 US milestone, including any designation of covered frontier models and details about developer participation. Agencies, contractors and AI companies will be watching for evidence that the voluntary evaluation window affects procurement or release decisions.
One day later, companies serving the European market will face the next EU AI Act application date unless the legislative process changes particular deadlines first. Enforcement guidance, national regulator activity and early compliance decisions will show how firmly the three regimes function as real deployment gates rather than formal requirements with limited reach.
Key Questions
What changed in China on July 15?
According to Thorsten Meyer AI, China’s anthropomorphic-interaction measures took effect, extending existing AI oversight to services such as companions and agents that imitate human interaction.
Does the United States require approval before an AI model is released?
Not under the framework described by the source. It provides a voluntary 30-day evaluation period, with classified testing and procurement-related incentives encouraging developers to participate.
Does the entire EU AI Act begin on August 2?
No. The law has entered application in stages since February 2025. August 2, 2026 is the next broad milestone, although proposed legislation could alter some high-risk-system deadlines.
Are the three pre-release systems equivalent?
No. China uses regulatory approval and continuing state oversight, the EU relies on conformity and risk-management obligations, and the US system described by the source uses voluntary national-security evaluation.
Do these regimes cover open-weight AI models released abroad?
Coverage remains uncertain and depends on where a developer operates, distributes or deploys a model. The source identifies foreign open-weight releases as a major gap because they may sit outside direct pre-release authority.
Source: Thorsten Meyer AI
