When the European Commission proposed a regulation on artificial intelligence in April 2021, it arrived with a title that seemed almost too tidy: the Artificial Intelligence Act. Three words, one acronym, and a promise of comprehensiveness that the 108-page text could not possibly deliver. The title suggested a single, coherent framework governing all of AI. The text, however, carved out exceptions for military systems, exempted most open-source models, and left entire categories of AI applications to existing product safety legislation. The gap between the name and the content was not an accident. It was the product of a deliberate, institutional process—one that reveals more about how EU policy works than most of the recitals that follow.

Legislative titles in the European Union are not neutral descriptors. They are framing devices, negotiated with nearly as much care as the articles they introduce. A well-chosen title can mobilize political support, attract media attention, and shape the interpretive lens through which courts and regulators will later read the text. A poorly chosen one can create expectations that the law cannot fulfill, providing ammunition for critics and confusion for implementers. The title problem, as I have come to think of it, is a microcosm of the broader policy implementation gap: the simplification necessary for public communication often obscures the trade-offs and technical detail that determine whether a law actually works.

To understand how titles acquire this power, it helps to trace their origins. The Commission’s Directorate-General responsible for a proposal typically drafts the initial title, often in consultation with the Secretariat-General and the Commissioner’s cabinet. This draft is not a mere label; it is a political signal. Calling something a “regulation” rather than a “directive” already communicates ambition and uniformity. Adding a short, memorable name—the “Digital Services Act,” the “Data Act,” the “Chips Act”—signals that this is not just another piece of technical legislation but a flagship initiative. The naming choice is strategic from the start, designed to position the proposal within the Commission’s annual work programme and to make it legible to journalists who will never read the full text.

Once the proposal reaches the Council and the European Parliament, the title becomes subject to the same trilogue negotiations as the rest of the text. Member states may object to a title that overpromises. The Parliament may insist on a name that reflects the rights-based framing it prefers. In the case of what became the Digital Services Act, early drafts carried more cumbersome titles referencing “a Single Market for digital services” and “harmonised rules.” The final, pithy name emerged only after months of negotiation, and it was chosen precisely because it sounded like a counterpart to the Digital Markets Act—a pairing that made the Commission’s digital agenda appear coherent and symmetrical, even though the two regulations address fundamentally different problems with different legal bases.

The consequences of these naming choices ripple outward in ways that are easy to underestimate. Consider the so-called “Right to Repair” directive. The title evokes a clear, individual entitlement: a right that consumers can assert against manufacturers. The actual text, adopted in 2024, is far more circumscribed. It obliges producers to repair certain products for a period after purchase, but only where repair is “technically possible” and only for product categories listed in an annex. It does not create an unconditional right. It does not cover all products. It does not even apply to all components. Yet the title “Right to Repair” has already shaped public expectations and activist demands in ways that the directive’s negotiators anticipated and, in some cases, encouraged. The name does political work that the articles cannot.

This dynamic is not unique to the EU, but the Union’s multilingual, multi-institutional structure amplifies it. A title must work in 24 official languages, each of which may carry slightly different connotations. The English “AI Act” sounds crisp and definitive; the French “règlement sur l’intelligence artificielle” is more descriptive and less branding-oriented. The German “KI-Verordnung” is bureaucratic shorthand. These variations matter because journalists, industry groups, and national parliaments often encounter the legislation first through its translated title. A name that sounds like a ban in one language and a framework in another can generate divergent political reactions before anyone has read the first recital.

The institutional machinery behind titling is largely invisible to outsiders. The Commission’s Better Regulation guidelines say nothing about how to name a legislative act. There is no interinstitutional agreement on titling conventions. The process is governed by precedent, political instinct, and the quiet work of communications officials who understand that a title is the first—and sometimes the only—part of a regulation that most citizens will ever encounter. In trilogues, titles are often settled last, after the substantive compromises have been reached, because the name must reflect the political deal without reopening the substantive disagreements. This sequencing means that titles are sometimes slapped onto texts they no longer accurately describe.

The AI Act is a case study in this mismatch. The title promises a horizontal regulation of artificial intelligence. The text delivers a product safety framework grafted onto a risk-classification system that exempts most AI applications and defers to sectoral legislation for many others. The “high-risk” category, which the title implies will capture the most dangerous systems, in practice covers a narrow set of use cases defined by an annex that can be amended through delegated acts. The title creates an expectation of comprehensiveness that the text deliberately avoids, because the Commission knew that a truly horizontal AI law would be impossible to negotiate. The name, in other words, was chosen to maximize political impact, not descriptive accuracy.

Savvy actors exploit this gap. Industry groups invoke the title to claim that a regulation covers their sector, even when the text explicitly excludes it. NGOs use the title to demand enforcement actions that the legal basis does not support. National regulators cite the title in guidance documents to justify expansive interpretations of their mandates. The title becomes a resource for institutional actors who know that most people—including many policymakers—will never read past it. This is not necessarily malicious; it is a rational response to a system in which the name carries more weight than the fine print.

The tension between memorability and accuracy that plagues EU legislative titling is not confined to Brussels. It is a general cognitive problem that anyone who has tried to name a book, a project, or even a policy paper will recognize. A title must be short enough to remember, distinctive enough to stand out, and faithful enough to the content that it does not mislead. These three desiderata are almost always in conflict. Short and memorable titles tend to oversimplify. Accurate titles tend to be long and forgettable. The EU’s solution—short, branding-friendly names that imply more coherence than the texts deliver—is one way to resolve the tension, but it comes at a cost.

This is the same balancing act that tools like a a novel title generator that fits the project attempt to automate for authors. Such generators, including those built into platforms like Reedsy’s writing studio, ask users to specify genre, core conflict, tone, and comparative titles, then produce options that range from the commercial to the literary. The Reedsy generator, for instance, lets writers choose between a “commercial mode” that favors broad market appeal and a “literary mode” that leans toward evocative, resonant language. Each suggestion comes with a one-line explanation of what it captures, making the trade-off between catchiness and fidelity explicit. The generator is not a replacement for human judgment—its creators are careful to note that it is “better at sparking a direction than landing the final answer”—but it makes visible the choices that EU legislative drafters make invisibly, behind closed doors, without ever acknowledging that they are making them at all.

The comparison is instructive precisely because of the differences. A novelist can afford a title that is evocative rather than descriptive; a regulation cannot. A novelist can change the title after the book is written; an EU legislative act, once published in the Official Journal, carries its name into every subsequent citation, court ruling, and implementation guideline. A novelist answers to readers; a regulation answers to courts, which may interpret the title as evidence of legislative intent. The stakes of getting the name wrong are higher in law than in literature, yet the EU has no systematic method for testing whether a title accurately reflects the text it introduces.

What would such a method look like? It would require, at minimum, a practice of revisiting titles after the substantive negotiations conclude, with an explicit check against the final text. It would require the institutions to ask: does this name still describe what we have actually agreed? It would require a willingness to change a politically attractive title when the content no longer supports it. None of this happens routinely. The political capital invested in a name like “AI Act” is too great to abandon, even when the text has drifted far from the original concept. The result is a growing archive of EU legislation whose titles promise more than the articles deliver, creating a standing invitation to disappointment and strategic misinterpretation.

The problem extends beyond flagship initiatives. Even technical regulations acquire names that shape their reception. The “Medical Devices Regulation” sounds like a comprehensive overhaul; in practice, it interacts with dozens of other texts and relies on standards developed by CEN and CENELEC that are not mentioned in the title. The “General Data Protection Regulation” is not, in fact, general: it exempts national security, applies differently to law enforcement, and leaves entire processing operations to member state law. The title “GDPR” has become so powerful that it now functions as a brand, invoked in contexts far beyond its legal scope. This is a success of political communication, but it is also a source of confusion that data protection authorities spend considerable resources trying to correct.

The Authors Guild, in its recently updated AI best practices for writers, articulates a related concern about the gap between surface and substance. The Guild warns that AI-generated text can appear coherent and authoritative while being “generic mashups of pre-existing works” that lack the “unique view and thoughts and unique voice” of a human author. The document is not about legislative titles, but its core insight applies: when a text—or a title—is produced through a process that prioritizes surface plausibility over substantive fidelity, the result can mislead even careful readers. The EU’s titling process, for all its political sophistication, sometimes produces names that are the institutional equivalent of an AI hallucination: plausible, memorable, and disconnected from the underlying reality.

This is not an argument for boring titles. A regulation called “Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence and Amending Certain Union Legislative Acts” would be accurate but unusable. The short name serves a genuine function: it makes the legislative agenda legible to citizens, journalists, and national parliaments. The problem is not the existence of short names but the absence of institutional mechanisms for ensuring that they remain tethered to the texts they represent. When the gap between name and content grows too wide, the title ceases to be a useful shorthand and becomes a source of systemic misunderstanding.

One way to narrow this gap would be to treat the title as a living document during negotiations, subject to revision at each trilogue milestone. Another would be to require the final legal text to include a “scope and limitations” clause that explicitly addresses what the title does not cover—a kind of truth-in-advertising requirement for legislation. A third would be to develop institutional memory about which naming strategies have produced the largest implementation gaps, so that future drafters can learn from past mismatches. None of these solutions is perfect, but all would acknowledge that the title problem is real and consequential.

The deeper lesson is about the relationship between simplification and governance. Democratic politics requires simplification: citizens cannot be expected to read 108-page regulations, and journalists cannot explain every recital. But simplification always involves selection, and selection always involves distortion. The question is not whether to simplify but how to manage the distortions that simplification introduces. The EU’s current approach—letting titles emerge from political negotiation without systematic quality control—manages these distortions poorly. It produces names that are optimized for launch-day press releases rather than for the decade of implementation that follows.

This is, in the end, a problem of institutional design. The Commission, the Parliament, and the Council all have incentives to produce memorable titles that attract positive coverage. None of them has a strong incentive to ensure that the title remains accurate after the legislative deal is done. The cost of inaccuracy is borne by implementers, courts, and citizens who encounter the gap between name and content only after the political spotlight has moved on. The benefit of a catchy name is captured immediately by the institutions that negotiate it. This asymmetry is a classic principal-agent problem, and it will persist until someone—perhaps the European Court of Auditors, perhaps the Joint Research Centre, perhaps a future interinstitutional agreement—takes responsibility for closing the gap.

Until then, readers of EU legislation would do well to approach every title with a measure of skepticism. The “AI Act” is not really an act on AI; it is a product safety regulation with a branding problem. The “Right to Repair” is not really a right; it is a limited obligation on producers. The “Digital Services Act” is not really about digital services in general; it is about intermediary liability and systemic risk management for very large platforms. The names are not lies, but they are not the whole truth either. They are the first move in a long game of framing, implementation, and interpretation—a game that begins before the first recital is read and continues long after the last article enters into force.

The title problem, in short, is not a trivial matter of nomenclature. It is a window into how the EU governs: through compromise, through communication, and through the constant tension between what can be said simply and what must be done complexly. Paying attention to titles—who writes them, who negotiates them, and what they leave out—is one way to become a more careful reader of policy. It is also a reminder that the most consequential governance often happens in the spaces between words, where the name promises one thing and the text delivers another, and where the real work of implementation must bridge the difference.