Global AI Governance Arena
Ten scenarios drawn from the rules actually in force in the European Union, the United States and China. Answer, then read why, with the source beside every claim.
Literacy arena
10 questions across three jurisdictions. Every answer, right or wrong, comes back with a short explanation and the source it rests on. This is a study aid, not an exam, and nothing you do here is graded or reported.
Ringside archive
The things that surprise practitioners most about the three regimes. Every card names its source.
The AI Act's top tier: €35M or 7% of global turnover for prohibited practices, beats even GDPR's €20M/4%, making it the toughest penalty regime in EU digital law. And GPAI model providers are policed directly by Brussels (the AI Office), not just national watchdogs.
The Act applies to any provider or deployer whose AI system's output is used in the EU: a recruiter in Texas or a vendor in Shenzhen screening EU candidates is in scope, and non-EU GPAI providers must appoint an EU authorised representative.
Emotion-recognition AI is banned outright in workplaces and schools (since Feb 2025), but used on shoppers or airline passengers it merely triggers a transparency notice. The AI Act regulates context of use, not technology, which is why ~85% of AI systems face no new obligations at all.
The TAKE IT DOWN Act (May 19, 2025) criminalizes nonconsensual intimate imagery: including AI 'digital forgeries', and a year later put the FTC in charge of 48-hour platform takedowns with penalties up to $53,088 per violation. There is still no general federal AI statute.
Colorado's landmark SB 24-205 (2024) was delayed once, sued by Elon Musk's xAI, opposed in court by Trump's DOJ, stayed by a federal judge on April 27, 2026, and repealed-and-replaced by a much weaker disclosure law (SB 26-189) on May 14, 2026, before a single day of enforcement.
While Congress debates, the FDA has authorized more than 1,200 AI-enabled medical devices (about three-quarters in radiology) through existing device-review pathways: more concrete algorithmic oversight than all US AI-specific laws combined. Meanwhile, the Senate killed a proposed 10-year ban on state AI regulation 99 to 1.
Unlike the US or EU, anyone can look up the CAC's filing lists: 868 generative AI services had completed national filing and 530 GenAI applications had registered with provincial CACs as of April 30, 2026, up from just 61 filings in 2023. Unfiled chatbots simply cannot legally operate in China.
The CAC had published 14 batches of deep-synthesis algorithm filings covering roughly 5,672 algorithms as of January 7, 2026, and filings can be cancelled too: 32 algorithms were deregistered in 2025 when products or companies ceased operations.
When the Anthropomorphic AI Interaction Measures took effect on July 15, 2026, ByteDance's Doubao and Alibaba's Qwen shut down their personalized companion features for hundreds of millions of users rather than comply with anti-addiction design, 2-hour break prompts, and the ban on virtual companion services for minors.
The syllabus
The 10 scenarios and the point each one makes, with the source behind it. The questions and their options stay in the quiz, so reading this first costs you nothing but the surprise.
GUESS THE REGIME: European Union
Emotion recognition in workplaces and schools is PROHIBITED in the EU since 2 Feb 2025 (Art. 5(1)(f)); in the US it draws bias-audit/civil-rights exposure; China has no dedicated employment-AI law.
TRUE OR STAMPED: United States
No comprehensive federal statute exists. Binding rules come from states (TRAIGA, SB 53…), cities (NYC LL144), and agencies (FDA, FTC). The Senate even killed a state-law moratorium 99 to 1.
Source: EO 14179: Removing Barriers to American Leadership in AI
NAME THE NUMBER: China
868 services + 530 registered applications: up from 61 in 2023. Unfiled chatbots can't legally operate, which is why ChatGPT is unavailable in China.
Source: S7: CAC GenAI filing registry
GUESS THE REGIME: European Union
EU Art. 5(1)(h) bans it except for missing persons, imminent threats, and serious-crime suspects, with prior authorization. The US has no federal rule; China permits state use under PIPL Art. 26 while restricting commercial use.
NAME THE NUMBER: European Union
€35M or 7% of worldwide turnover: bigger than GDPR's ceiling. GPAI providers are policed directly by the Commission's AI Office.
TRUE OR STAMPED: United States
It never took effect: sued by xAI (Apr 9, 2026), opposed by Trump's DOJ, stayed by a federal court (Apr 27), then repealed-and-replaced by the weaker SB 26-189 (May 14, 2026).
Source: SB 24-205: Colorado AI Act
GUESS THE REGIME: China
China's GenAI Interim Measures (Art. 17): pre-launch filing since Aug 2023. The EU only requires disclosure that you're talking to an AI (from 2 Aug 2026); the US has no general chatbot law.
Source: GenAI Measures: Interim Measures for Generative AI Services (CAC Order No. 15)
TRUE OR STAMPED: United States
1,247+ AI-enabled medical devices authorized (~76 to 80% radiology): more concrete algorithmic oversight than all US AI-specific laws combined.
GUESS THE REGIME: China
China's Anthropomorphic AI Measures took effect 15 Jul 2026 (2-hour break prompts, minor companion ban); ByteDance's Doubao and Alibaba's Qwen pulled companion features at the deadline.
Source: Decree 21: Interim Measures for AI Anthropomorphic Interaction Services
NAME THE NUMBER: European Union
~85%: spam filters, game AI, recommendation engines. The AI Act regulates context of use, not technology.
Literacy is the one duty every regime agrees on. The EU wrote it into Article 4.
A study aid, not legal advice. Every question rests on a published source you can open from the citation chip beside the explanation. Nothing you answer is stored on our servers; the last score chip lives in your browser alone.
Ten scenarios tell you whether you can place a rule in the right jurisdiction. The programs teach the work that follows: classification calls, conformity assessment, filings, documentation, and the judgment to defend every decision.