MiCA and the AI Act: Europe Exports Process, Imports the Frontier
Two thick legal objects, published thirteen months apart. MiCA for crypto-assets, the AI Act for artificial intelligence.[2][5] Both carry the same European signature: harmonize the market, classify the risk, fine the turnover, export the PDF.
MiCA and the AI Act are not the same statute. They rhyme so hard that treating them as coincidences is a failure of pattern recognition. Together they are the continent’s theory of power in the 2020s: if we cannot always build the frontier, we will specify the conditions under which the frontier may be sold here. Sometimes that is wisdom. Sometimes it is a lifestyle brand with annexes. I have already made that case for the AI Act alone.[1] This is the twin portrait.
What MiCA and the AI Act are for
MiCA is Regulation (EU) 2023/1114. It aims at market integrity for crypto-assets in the Union: issuers, service providers, disclosure, authorization, consumer-facing conduct. Its taxonomy is narrower than the coverage suggests. MiCA creates no legal category called “stablecoin”. It creates asset-referenced tokens, which purport to hold a stable value by referencing another value or right, and e-money tokens, which reference a single official currency, and then everything else, which falls under the general regime and its crypto-asset white paper. Service firms are crypto-asset service providers, and nobody may provide those services in the Union without authorization.[2] It is a financial-market statute wearing digital clothes.
Honest firms get a passport logic: authorize once, tell your home authority which Member States you intend to serve, and start cross-border within fifteen calendar days of the notification.[3] Scam shops get a thinner fog to hide in, at least inside the regulated perimeter.
The AI Act is Regulation (EU) 2024/1689. It aims at product safety and fundamental rights: outright prohibitions, a conformity process for the high-risk uses listed in Annex III, a separate chapter of duties for general-purpose AI models, and transparency obligations that require, among other things, that synthetic audio, image, video and text be marked in a machine-readable format.[5] It is a product statute wearing ethics clothes. Honest deployers get a compliance industry. Rights-hostile product categories get bans on paper.
Both statutes bite the same way: administrative fines scaled to worldwide turnover, up to 7% for a prohibited AI practice and up to 3% for a transparency failure, with MiCA leaving the equivalent arsenal to national authorities.[8] And both reach for the Brussels Effect: write rules others must swallow if they want EU customers. GDPR proved the pattern can work. Working is not the same as winning the industries the rules describe.
Who they actually discipline
Same cast as always.
| MiCA-ish reality | AI Act-ish reality | |
|---|---|---|
| Named firm with EU customers | Hard to ignore | Hard to ignore |
| Offshore scam / anonymous shop | Plays jurisdiction games | Plays jurisdiction games |
| Self-hosted user / open stack | Edge cases and politics (unhosted fights) | Open weights and local models (weights) |
| Foreign industrial policy | Not aimed there | Not aimed there |
The rhyme is the point. Europe is excellent at building process that bites letterheads. It is weaker at building the letterheads that own the frontier: the model labs, the liquidity venues that set global prices, the chip and energy stack under both.
Why the twin PDFs still make sense
Fragmented national crypto rules were a gift to arbitrage and a curse to compliance teams. Fragmented AI rules would have been worse. Citizens deserve limits on biometric theater and on fraud dressed as a token. Single-market clarity is a public good for builders who want to sell once and ship. Trust can attract capital that fears the Wild West. Draghi told the Commission the Union adds regulatory burdens that are “self-defeating for those in the digital sectors” and that it needs at least 750 to 800 billion euro a year of additional investment, some 4.4 to 4.7% of 2023 EU GDP, to close the gap.[9] Neither half of that diagnosis requires nihilism about law.
A continent that remembers twentieth-century industrial disasters is allowed to be suspicious of “move fast.”
The failure mode they share
Rules as substitute for production. Announcing AI factories while the frontier training runs happen elsewhere: since 2023, every model at the top of Epoch AI’s capability index has been developed in the United States, which also holds roughly three quarters of the AI supercomputer performance in Epoch’s dataset against China’s 15%.[10] In 2024, US institutions released 40 notable AI models, China 15, and all of Europe 3; the next edition of the same index counted 50, 30 and 2 for 2025.[11] Policing token disclosures while dollar liquidity and corporate treasuries set the tempo of crypto. Demanding machine-readable marks on cooperative models while open weights and foreign APIs still run the interesting workloads.[1]
Also: theater layers. White papers nobody reads. Transparency badges nobody believes. Both statutes have hard cores worth enforcing (fraud, prohibitions, high-stakes process). Both will grow soft shells of compliance cosplay if supervisors sleep.
August 2026 was a good month to test that. The AI Act’s transparency duties became applicable on 2 August 2026, on the original schedule.[6] The Annex III high-risk regime, the part with the expensive teeth, did not. Six days before that deadline, the Digital Omnibus entered into force and moved stand-alone high-risk obligations to 2 December 2027 and product-embedded ones to 2 August 2028.[7] The labels arrived on time. The conformity assessments slipped by sixteen months. MiCA, for its part, has been fully applicable since 30 December 2024, and its token titles since 30 June 2024.[4] Which half of a statute survives contact with industry tells you what the statute was mostly made of.
Neither instrument is the opposite of the United States or China. They are Europe’s chosen instrument. The US leans capital and voluntary frameworks. China leans industrial policy plus speech control. Europe leans codification. Codification without capacity is dependency with better typography.
Crypto-AI hybrids will feel both
On-chain agents, trading bots, scoring oracles, “AI-managed” funds marketed to retail: the interesting products will touch both rulebooks. That is a feature for lawyers. It is a warning for builders. Design for two supervisors or design outside the perimeter on purpose. Pretending you are only a smart contract or only a chatbot is how you get surprised.
What Europe exports, and what it imports
Keep the hard cores. Enforce fraud and rights-hostile systems without apology. Stop confusing the thickness of the Official Journal with sovereignty.
Two rulebooks, both in force, both real. The frontier is still imported. The next European project is not a third PDF. It is energy, chips, capital formation, and labs that make the annexes apply to something built here.
- The Badge Is Not the Ethics; The AI Act Regulates Invoices. Open Weights Don’t Send Any..
- Regulation (EU) 2023/1114 on markets in crypto-assets (MiCA), OJ L 150, 9.6.2023. Subject matter: Art. 1. Definitions of asset-referenced token, e-money token and crypto-asset service provider: Art. 3(1)(6), 3(1)(7), 3(1)(15). Crypto-asset white paper: Art. 6 (other crypto-assets), Art. 19 (asset-referenced tokens), Art. 51 (e-money tokens). Authorization requirement: Art. 59.
- MiCA, Art. 63 (granting or refusal of authorization) and Art. 65 (cross-border provision of crypto-asset services): the home authority forwards the notification within 10 working days and the provider may start on receipt of that communication or 15 calendar days after submitting the information, whichever is earlier.
- MiCA, Art. 149: the Regulation applies from 30 December 2024, with Titles III (asset-referenced tokens) and IV (e-money tokens) applying from 30 June 2024.
- Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (AI Act), OJ L, 12.7.2024. Prohibited practices: Art. 5. High-risk classification: Art. 6 and Annex III. General-purpose AI models: Chapter V, Arts. 51 to 56. Transparency, including machine-readable marking of synthetic content: Art. 50.
- AI Act, Art. 113: general application from 2 August 2026; Chapters I and II from 2 February 2025; Chapter III Section 4, Chapter V, Chapter VII, Chapter XII and Art. 78 from 2 August 2025, except Art. 101; Art. 6(1) from 2 August 2027.
- Regulation (EU) 2026/1744 of 8 July 2026 amending Regulation (EU) 2024/1689 as regards simplification of the implementation of harmonised rules on artificial intelligence (Digital Omnibus on AI), published in the OJ on 24 July 2026 and in force from 27 July 2026. It defers Annex III high-risk obligations to 2 December 2027 and Annex I product-embedded high-risk obligations to 2 August 2028, leaving the Art. 50 transparency date untouched.
- AI Act, Art. 99: up to 35,000,000 euro or 7% of total worldwide annual turnover for breaching the Art. 5 prohibitions, up to 15,000,000 euro or 3% for other obligations including Art. 50, up to 7,500,000 euro or 1% for supplying incorrect information. MiCA, Art. 111, requires Member States to give competent authorities administrative penalties including fines, public statements, cease-and-desist orders and management bans.
- Mario Draghi, The future of European competitiveness, Part A: A competitiveness strategy for Europe (September 2024). Foreword, on adding regulatory burdens that are “especially costly for SMEs and self-defeating for those in the digital sectors”; section 5, Financing investments, for the minimum annual additional investment of 750 to 800 billion euro, 4.4 to 4.7% of EU GDP in 2023; and the Simplifying rules section on the growing regulatory burden and fragmented GDPR implementation.
- Epoch AI, Trends in AI Supercomputers (arXiv:2504.16026, April 2025): the United States accounts for about three quarters of global AI supercomputer performance and China for 15%, as of March 2025, in a dataset the authors estimate covers 10 to 20% of the global aggregate. Frontier leadership claim from Epoch AI, The Geopolitics of AI.
- Stanford HAI, AI Index Report 2025, Chapter 1, on notable AI models by geography in 2024 (United States 40, China 15, Europe 3); 2025 figures as reported by IEEE Spectrum from the 2026 AI Index (United States 50, China 30, Europe 2).