TL;DR
China’s anthropomorphic AI rules took effect July 15, followed by scheduled US frontier-model measures on August 1 and full application of the EU AI Act on August 2. The three systems share an interest in pre-release scrutiny but apply different tests and leave gaps around foreign open-weight releases.
China’s anthropomorphic AI rules took effect July 15, beginning a 19-day sequence in which the United States and European Union are also scheduled to tighten oversight of advanced AI systems before public deployment. The deadlines matter because they establish three different regulatory tests for developers operating across the world’s largest AI markets.
China’s Interim Measures for AI Anthropomorphic Interaction Services, issued in April by five agencies, apply the country’s existing approval model to companion AI, human-like interfaces and agents. The framework involves the Cyberspace Administration of China, the national development and industry ministries, public security authorities and market regulators.
China already requires covered public generative AI services to undergo security review and algorithm registration. According to the source material, regulators can seek design changes before registration, demand incident reports within 24 hours and order later adjustments. The new measures extend that state-supervised approach to systems designed for sustained, human-like interaction.
On August 1, the US framework described under Executive Order 14409 is scheduled to harden around a classified benchmark, National Security Agency designations and an optional 30-day government evaluation window. The EU AI Act is due to become fully applicable on August 2, adding risk classification, conformity checks, technical records and post-market monitoring. Unlike China’s system, the US process remains voluntary; unlike the US model, the EU framework creates broad legal duties for covered providers and deployers.
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.
Three Regulators, Three Tests
The clustered dates show a shared regulatory judgment that certain AI systems should face government scrutiny before broad release. They do not represent agreement on what that scrutiny should measure. China focuses heavily on content controls and social stability, the EU on fundamental rights and product safety, and the US framework on national security.
That difference has direct consequences for developers. A model approved or evaluated in one jurisdiction may still require separate testing, records and deployment controls elsewhere. Release calendars, model documentation and incident-response systems must now account for jurisdiction-specific architecture, rather than treating compliance as one reusable global review.
How the Deadlines Converged
China has operated a pre-deployment regime for public generative AI services since 2023. Its latest measures apply the same targeted, use-case-based regulatory method to anthropomorphic systems, including AI products that imitate human personality or maintain ongoing relationships with users.
The EU AI Act has entered force in stages. Its first prohibitions applied in February 2025, while obligations for general-purpose AI models began the following August. The August 2, 2026 milestone brings most of the remaining framework into application. The US has taken a narrower path, using access to government evaluation and potential trusted-partner status as incentives instead of requiring general pre-market approval.
“The convergence is real at the level of instinct. It is not convergence at the level of design.”
— Thorsten Meyer AI
Open Models Remain a Gap
It is not yet clear how effectively any of the three systems can reach open-weight models released abroad, especially when the developer lacks a local entity or does not seek government contracts. The source argues that this is a structural gap, but enforcement will depend on distribution channels, local deployers and jurisdictional reach.
The EU timetable also faces a pending legislative caveat. A Digital Omnibus proposal approved by the European Parliament on June 16 would move some high-risk deadlines, according to the source material. It has not yet completed Council adoption or Official Journal publication, so August 2 remains the operative date unless that process is completed. Which duties would move, and when, remains unsettled.
August Deadlines Drive Compliance
Developers operating in the United States must watch for NSA model designations, access conditions and details governing the voluntary evaluation window before August 1. Because the benchmark is classified, outside researchers and companies may receive limited visibility into how covered systems are judged.
In Europe, providers and deployers must work from the current August 2 timetable while monitoring the Digital Omnibus process. Regulators’ enforcement guidance, treatment of foreign open models and early compliance actions will show whether the three gates produce meaningful pre-release controls or mainly add obligations for companies already equipped to meet them.
Key Questions
What changed in China on July 15?
China’s anthropomorphic interaction measures took effect, extending regulatory oversight to human-like AI services such as companions and agents.
Does the United States require approval before an AI model is released?
No general approval requirement is described in the source. The US system offers a voluntary 30-day evaluation window tied to classified national-security criteria and possible trusted-partner benefits.
Is the EU AI Act deadline still August 2?
Yes, based on the current legal timetable. A pending Digital Omnibus proposal could move certain high-risk deadlines, but those changes are not yet in force.
Do the three regimes evaluate the same risks?
No. China emphasizes content and social stability, the EU centers on rights and product safety, and the US framework targets national-security concerns.
Are open-weight AI models covered?
Coverage depends on where and how a model is released or deployed. Foreign open-weight releases may sit beyond direct regulatory reach, although local distributors and deployers can still face obligations.
Source: Thorsten Meyer AI
