Three regimes, one week — the map of AI law is now drawn

Europe switched on enforcement, Beijing issued fines within days of its own deadline, and Washington's unifying bill sat stalled. The regulatory world stopped being theoretical and settled into three incompatible shapes.

Within a fortnight: the AI Act's Article 50 transparency duties began applying across the EU, and China fined twelve companies 4.2 million yuan in the first week of enforcing its companion-AI rules. Meanwhile the American preemption fight stayed stuck. Three approaches, all live, all mutually unaligned.

Beijing chose speed and picked its target

China's rules took effect on 15 July and penalties landed almost immediately. More telling than the tempo is the choice of first target: software that simulates emotional relationships with users. Of everything a regulator might have moved on, Beijing went at anthropomorphism — a judgement that the psychological surface of AI is where near-term harm concentrates.

Europe chose scope and inherited an evidence problem

Article 50's duties are easy to state and awkward to discharge. Proving you disclosed AI interaction and labelled synthetic media across a real product estate requires knowing what your systems emit, when, and through which surfaces. Compliance departments are discovering they need model-attestation tooling — which is why a research agenda suddenly has a paying customer.

Washington chose neither

The American vacuum is itself a policy. Every month the federal bill idles, more state statutes take effect and the patchwork it was meant to erase gets harder to erase. And the one thing holding the technical picture together across all three regimes is a report backed by thirty-plus countries that describes the risks without binding anyone to do anything about them.

Shared diagnosis, divergent enforcement. Multinational compliance teams will spend the next several years living in exactly that gap.

Bird & Bird — China's new regulations on AI anthropomorphic interactive services → · Cubbbix — AI regulation news August 2026 →