The National AI Governance Playbook · Part III · Chapter 8
Choosing the legal vehicle
Only one link of the chain needs a legal vehicle: the binding instrument. This chapter sizes the instrument to that link, compares the three vehicle families visible in public practice, and sets out what any vehicle must carry. The right choice depends on the jurisdiction, and combined routes are common.
By Myriam Ayada · MindXO · July 2026
In brief
Chapter 5 scoped the question: of the six links in the policy and regulatory chain, only link three, the binding instrument, is a matter of legislation. The rulebooks below it draw their force from it, and the standards beside it bind by citation, so the instrument itself can stay small. Public practice shows three vehicle families: a dedicated AI law, targeted amendments to existing statutes, and executive instruments under existing powers. Each trades speed against permanence and scope, none is correct in general, and the choice follows the jurisdiction's legal tradition, its legislative bandwidth and what the rollout needs bound first. Combined routes are common, and whatever the family, six contents must travel in the instrument: definitions, allocation, powers, the citation hook, funding authority and a review clause.
What actually needs to bind
Chapter 5 closed with a scoping observation: only link three of the chain, the binding instrument, raises the question of legislation. The regulatory rulebooks at link four draw their force from whatever binds above them, and the standards at link five bind by citation, when a rulebook or a contract references them. Nothing else on the chain needs a legal vehicle. The question this chapter answers is therefore narrower than it looks, and the answer depends on the jurisdiction. The playbook holds throughout that no vehicle is correct in general, and this chapter is where that position does its work.
Sizing the instrument to the link keeps the statute small. Six contents genuinely require binding force: definitions, so that every rulebook below the instrument means the same thing by the same words; the allocation of the four functions of Chapter 3 to named entities; powers to inspect and to sanction; the citation hook, the clause that lets rulebooks reference standards and assurance schemes without re-legislating; funding authority for the safety and security capability of Part IV; and a review clause with a date. Everything else can live lower on the chain, in rulebooks, guidance and licence conditions that adapt sector by sector, and in standards that revise on a technical cadence, where amendment does not require a legislature.
Three vehicle families
Public practice shows three families, and a design can combine them. The omnibus route writes a dedicated AI law; Regulation (EU) 2024/1689 is the reference case. The targeted route inserts AI provisions into the sectoral, data and consumer statutes the jurisdiction already has. The executive route reaches binding force through decrees, regulations under existing powers and statutory national plans; the United Kingdom's regulator-led approach approximates this family without a new statute. The cards compare the three on the four properties a design phase has to weigh.
V1. A dedicated AI law: the omnibus route
Speed: The slowest family. Regulation (EU) 2024/1689 was adopted four years after the white paper that opened the file, a representative interval for a major statute. Permanence: The strongest. A statute survives changes of government, and the same rigidity makes it slow to correct. Scope: Comprehensive. One instrument carries the whole design in one place, and adoption reads as a durable commitment at home and abroad. What it asks of the calendar: A major slot in the legislative program, in competition with every other priority of the term.
V2. Targeted amendments: riding the existing statute book
Speed: Faster. Provisions travel inside bills already moving, and existing supervisors and definitions carry them once passed. Permanence: Statutory, and therefore durable, though the provisions spread across several acts with separate amendment histories. Scope: Sectoral by construction. Coverage grows statute by statute, and it risks patchwork unless the coherence frame of Chapter 6 holds the pieces together. What it asks of the calendar: Modest per amendment. The cost is negotiating several smaller files across several ministries in place of one large one.
V3. Executive and existing statutory instruments: force under existing powers
Speed: The fastest binding force available. Decrees, regulations under existing powers and statutory national plans issue on the executive's own calendar. Permanence: The weakest. What one administration issues, the next can withdraw, so this family depends most on later consolidation. Scope: Bounded by the powers already conferred. The route reaches only as far as the enabling statutes do. What it asks of the calendar: Close to none, which is the family's appeal and the reason designs that start here usually plan a statutory follow-on.
Sheet 08 of 13, three vehicle families and what every vehicle must carry: a dedicated AI law, targeted amendments and executive instruments compared across four properties, speed, permanence, scope and calendar demand. The dedicated law is the slowest, the most durable, the broadest and the heaviest ask of the calendar; targeted amendments sit between on every property; executive instruments are the fastest, the least permanent, the most bounded and the lightest ask. Beneath the comparison, a strip lists the six contents every vehicle must carry: definitions, allocation, powers, the citation hook, funding authority and a review clause.
Choosing among them
Five criteria do most of the sorting. Legal tradition comes first: civil law and common law jurisdictions reach binding force through different default instruments, and a vehicle that is routine in one tradition may have no convenient counterpart in the other. Legislative bandwidth follows, the realistic number of major files a legislature passes in a term. Sequencing is third: Chapter 7 settled what the rollout needs bound early and what can wait, and the vehicle has to deliver force on that schedule. Fourth is the existing statute book, since a data protection law or a strong sectoral act can often carry AI duties by amendment. Fifth is the review cycle the design can realistically get, because an instrument that cannot be revisited on a known date ages quickly in a field that moves on model releases.
A combined route is common in practice. An early executive instrument anchors the allocation and the capability funding while the machinery is built, and consolidating legislation follows once the regime has operating evidence. The public exemplars mark the ends of the range: the European Union carried one omnibus file from the 2020 white paper to the 2024 regulation, and the United Kingdom has so far held its design in regulator guidance under existing powers, with statutory consolidation kept open. Where a given jurisdiction should sit on that range is a property of the jurisdiction, of its tradition, its calendar and its statute book, and of nothing more general.
Common failure mode. The omnibus reflex. A dedicated AI law is chosen because it signals seriousness, and the design then waits years for a legislative slot while holding no force at all. The inverse failure anchors nothing, so the design stays administrative and lapses at the first change of government. The working alternative is to size the instrument to link three and to combine vehicles across time: an early instrument for force now, consolidation for permanence later.
What every vehicle must carry
Whatever the family, six contents must travel in the instrument. They mirror the sizing of the first section, because they are the parts of the design that nothing lower on the chain can supply for itself.
- Definitions aligned to the standards the rulebooks will cite, so the words in the statute and the words in the test carry the same meaning.
- Allocation naming the entities that hold each of the four functions of Chapter 3.
- Powers to inspect and to sanction, the part of the design only a binding instrument can create.
- The citation hook, so rulebooks can reference standards and assurance schemes, and both can revise without reopening the statute.
- Funding authority for the safety and security capability, so the horizontal regime of Chapter 4 is financed by the same instrument that relies on it.
- A review clause with a date, so revision arrives on a schedule the field's pace justifies.
For parties to the Council of Europe Framework Convention on Artificial Intelligence, accession adds one further consideration. The convention binds at international law and still requires domestic carriage before anything changes inside the jurisdiction, the citation-without-transposition pattern of Chapter 1. Whichever vehicle the design chooses becomes that carriage, and the allocation and powers it carries are what give the convention's obligations domestic effect.
The vehicle settles how the design acquires force. A separate question is who steers the chosen instrument through the machinery, holds the drafting pen, defends the calendar slot and answers for the design between ministries. Part III closes with that question: Chapter 9 asks who leads the design phase.
Three questions for every government
- Which contents of the design need binding force now, and which can bind later or lower on the chain?
- Which existing statutes could carry targeted AI provisions, and what would be lost against a dedicated law?
- What is the realistic legislative calendar, and what interim instrument holds the design until it clears?
Selected public sources
- Regulation (EU) 2024/1689 (AI Act), European Union, 2024
- Framework Convention on Artificial Intelligence, Council of Europe, 2024
- A pro-innovation approach to AI regulation, white paper and government response, United Kingdom, 2023 and 2024
- White Paper on Artificial Intelligence: a European approach to excellence and trust, European Commission, 2020