TL;DR
China’s rules for human-like AI services took effect on July 15, while separate US and EU measures are due on August 1 and August 2. The three systems share an interest in examining advanced AI before public deployment, but differ sharply on approval, enforcement and policy goals.
Three major AI jurisdictions are activating separate pre-release controls within 19 days: China’s measures for human-like AI services took effect on July 15, a US evaluation framework is due to harden on August 1, and the EU AI Act reaches its scheduled full-applicability date on August 2. The concentrated timetable matters because developers operating across markets will face different tests for the same model or service.
China’s Interim Measures for AI Anthropomorphic Interaction Services, issued by five state agencies, extend the country’s existing approval structure to systems designed to simulate human interaction, including AI companions and agents. According to the source dispatch, covered services may face security reviews before deployment, algorithm registration and regulator-directed changes.
In the United States, the dispatch describes the August 1 milestone under Executive Order 14409 as a voluntary 30-day access window. Participating frontier-model developers would provide government evaluators with pre-release access under classified criteria, while trusted-partner status and procurement access would supply incentives rather than a legal approval requirement.
The EU AI Act is scheduled to become fully applicable on August 2 after a staged rollout that included prohibited-practice rules in February 2025 and obligations for general-purpose AI models in August 2025. Its system centers on risk classification, conformity reviews, technical records and monitoring after products enter the market.
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 AI Legal Handbook: A Guide to the Laws of Artificial Intelligence and the Future of Regulation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
One Model, Three Regulatory Tests
The timing shows a shared policy instinct: some advanced AI systems should face government scrutiny before reaching users. The resemblance largely ends there. China focuses on content control and social stability, the EU emphasizes fundamental rights and product safety, and the US framework described by the dispatch concentrates on national-security risks.
For developers, jurisdiction is becoming part of product architecture and release planning. A company may need separate documentation, evaluation access, incident procedures and technical controls for each market. Smaller laboratories could face a heavier relative burden because multi-market compliance costs are easier for large companies to absorb.
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.
How the Three Systems Differ
China has required security reviews and algorithm filings for public generative-AI services since 2023. The new measures apply that model to human-like interaction services, with the state retaining authority over registration, incident reporting and requested algorithm changes. The dispatch characterizes this arrangement as the state acting as a co-designer.
The EU uses a broader product-regulation model based on risk categories and conformity duties. The US approach remains lighter: it creates an evaluation channel for participating developers but, based on the supplied account, does not give the government a general power to block release. The United Kingdom remains outside this convergence, retaining a sector-led, principles-based system without a formal pre-release gate.
“Same instinct, three theories of a gate.”
— Thorsten Meyer AI dispatch
Software Testing with Generative AI
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. The source says a Digital Omnibus proposal, approved by the European Parliament on June 16 by 423 votes to 57, with 174 abstentions, would delay some high-risk-system deadlines. It has not yet completed Council adoption or Official Journal publication, so the dispatch treats August 2 as the operative date.
Questions also remain about US participation rates, the classified benchmark and which frontier models the National Security Agency will designate as covered. None of the three systems clearly closes the gap created by open-weight models released outside its jurisdiction, and the practical reach of enforcement across borders remains uncertain.
The AI Legal Handbook: A Guide to the Laws of Artificial Intelligence and the Future of Regulation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
August Deadlines Put Rules to Test
Attention now moves to the August 1 US milestone, including any covered-model designations and evidence that developers have joined the voluntary access program. In Europe, companies and regulators will watch for Council action on the Digital Omnibus before the AI Act’s August 2 deadline.
China’s implementation will provide the first evidence of how the new rules operate in practice, including whether authorities demand service changes, new filings or incident reports. Enforcement guidance and early regulatory decisions will show whether the 19-day convergence produces meaningful pre-release controls or mainly three separate compliance tracks.
Key Questions
What changed in China on July 15?
China’s anthropomorphic-interaction measures took effect, extending regulatory controls to AI companions, agents and other services designed to imitate human interaction.
Can the US government block a model under this framework?
Based on the supplied account, the central US program is a voluntary evaluation window, not a general approval system. Participation may affect trusted-partner or procurement status, but the exact consequences remain unclear.
Does the full EU AI Act apply on August 2?
August 2 remains the scheduled legal date in the source material. A pending Digital Omnibus measure could move some high-risk deadlines, but it would require final adoption and publication before changing the law.
Do the three regimes evaluate the same risks?
No. China emphasizes content and social-stability controls, the EU focuses on rights and product safety, and the US framework emphasizes national-security evaluation.
Are open-weight AI models covered?
Coverage is incomplete. The dispatch says open-weight releases from outside a jurisdiction may bypass these gates, leaving a major question about cross-border enforcement.
Source: Thorsten Meyer AI
