Quién escribe las reglas de la IA: el poder detrás del vacío

There's a question that doesn't appear in World Economic Forum reports, Congressional hearings, or academic analyses about the future of work. Not because it's unthinkable, but because asking it out loud forces us to name what many prefer to leave implicit: who writes the rules and whose interests they serve.

The public debate about artificial intelligence and employment has organized itself around answers to a different question. Universal basic income as a safety net against automation. Alternative economic models to the market. Neo-Luddism as active rejection. These are serious proposals with solid arguments. They all address what model we want. None tackle the prior question: who has the authority to decide, and through what processes do they exercise it.

This is not a technical question. It's a question of power. And power doesn't wait to be formulated elegantly.

The dominant view claims that the State already provides answers. The European Union's AI law is considered the most ambitious regulation available. It classifies systems by risk, demands transparency and audits. On paper it represents progress. There are researchers who defend it as an exportable model and they're right on several points: it requires human oversight and documentation in high-risk applications like credit, employment, and critical infrastructure.

The problem isn't what the regulation says. It's what it avoids touching.

The law includes no provision about who decides the direction of technological innovation. It regulates transparency, not power. It regulates risk, not distribution. Meanwhile, subsequent measures have already pushed in the opposite direction by diluting high-risk obligations and softening compliance requirements. Regulation arrives as risk management, not as a dispute over who's steering the ship. These are two different matters, though they share vocabulary.

While institutions classify, companies write. And they do so with names and surnames.

The network of political action committees grouped as Leading the Future reported over fifty million dollars to electoral authorities. Two of its main backers contributed twenty-five million each. The same strategist who coordinated similar efforts in other fields leads this network. Its stated objective isn't to promote good artificial intelligence. It's to block state regulation, starting by opposing specific legislators who pushed safety laws. This isn't diffuse lobbying. It's a precise operation to keep empty the space where someone might ask the question that no one gets paid to pose.

This reveals who holds the pen by default when no one else claims it.

There's a detail that illustrates something structural. The report on job losses from automation published by Senator Bernie Sanders' office in 2025 used projections generated by ChatGPT in part of its work. The instrument modeling the damage is the same one producing it. This doesn't speak to hypocrisy. It shows how the tools of power infiltrate even resistance to power. Whoever wants to dispute the terms must be conscious of what they're disputing them with.

What's curious is that real change, when it comes, rarely emerges from the front of the train. The Snowpiercer image remains useful: in a closed system, terms were never rewritten through voluntary concession. They always required sustained pressure from the back, car by car. Actors' unions maintained prolonged strikes until obtaining concrete protections on AI use and digital replicas. Longshoremen negotiated direct prohibitions on fully automated cranes. Writers incorporated authorship safeguards. None of these victories emerged from the State or academic papers. They came from workers who understood that collective bargaining remains, for now, the mechanism that actually disputes terms in real time.

That negotiation functions as a de facto regulator while institutions are still classifying risks. Brynjolfsson has pointed this out clearly: when machines substitute for labor, workers lose not just income but economic and political bargaining power. They become dependent on those who control the technology. That, without euphemisms, is exactly the question no one gets paid to ask.

It's not yet resolved how to scale this up. The achievements protect sectors with high organization and visibility. They don't yet cover mid-level accountants, junior analysts, or call center operators in different cities. Voices disputing from emerging economies remain under-articulated, not for lack of ideas but because the forums where decisions are made exclude them by design. There are aspects I don't fully understand. I continue exploring how a broader dispute might be articulated.

One fact complicates catastrophist versions: around ninety-five percent of companies that have implemented generative AI still report net losses on that investment. The massive transformation of employment that some reports model isn't happening at the speed of headlines. That doesn't mean it won't happen. It means there's time. And that time can be used to dispute terms or cede them through omission.

The closing question isn't what model we want. That one already has too many competing answers.

Who's going to hold the pen when the rules are written, and what are you doing to ensure it's not just whoever already holds it?

1. European Parliament. EU Artificial Intelligence Act (2024). Official text and analysis of high-risk provisions.

2. Federal Election Commission. Leading the Future PAC — Financial Disclosures (2024-2025). Public filing, over fifty million dollars reported.

3. Brynjolfsson, Erik & McAfee, Andrew. The Second Machine Age. W.W. Norton, 2014. Chapters on bargaining power and technological distribution.

4. SAG-AFTRA. Television and Film Agreement 2023 — AI Provisions. Official union statement.

5. International Longshoremen's Association. USMX-ILA Framework Agreement 2024 — Port equipment automation clause.