Madrid Bar Association seeks penal reform for legal AI

ICAM pushes to amend Incivil Code to cover unlicensed legal AI

Could a chatbot that drafts insurance claims better and cheaper than a local firm end up facing incivil charges? The Madrid Bar Association (ICAM) has registered a proposal with the Spanish Congress to modify Article 403 of the Incivil Code, which covers professional intrusion, to adapt it to platforms, chatbots, and automated systems that offer public "consulting" or legal advice without direct and effective professional supervision. The proposal does not seek to regulate technology: it aims to update the incivil offense.

Eugenio Ribón, Dean of the ICAM, justifies this with an argument that sounds reasonable until read twice. He states that qualification and bar membership are "not corporate formalities, but institutional mechanisms guaranteeing the public interest." He warns that Article 403 was designed for "a person practicing without a degree," not for a product sold massively whose authorship is diluted "in technological structures." In short: the intruder no longer has a face, it has an API.

What exactly does the ICAM propose?

The initiative starts from a specific premise: digital transformation allows simulating legal guidance at scale and displacing professionals. The association speaks of "disintermediation of the professional-client relationship" and "functional substitution of human professional judgment." The risk they highlight is not that AI informs: it is that citizens sign, appeal, renounce rights, or accept agreements based on automated responses lacking ethical duties or equivalent professional liability.

The ICAM admits that Article 403 works against identifiable human intruders but may fall short when services arrive through platforms with "diffuse organizational authorship." Therefore, the proposal does not target users who ask questions: it targets those who commercialize the system. This is a huge legal difference and explains why the debate has become so heated.

The guild's argument vs. the cost counterargument

The corporate defense relies on guarantees: training, ethics, and liability. The counterargument circulating in the discussion is more prosaic. Part of the analysis holds that much of the legal advice provided by small firms consists of filling out templates with client data, and that a well-trained language model does this with fewer errors. The real dispute is not about error, but disintermediation: if the client solves it alone, someone loses the fee.

The figures cited in the conversation are telling. Some estimate a minimum of €80 per document for a lawyer's invoice for an administrative claim, while others claim to have processed several such claims using generative models and won them. None of these experiences are verifiable, but they set the tone: the perception that the cost of professional intermediation no longer stands on its own.

The counterattack: bad questions lead to losses

The strongest reply to enthusiasts is practical, not legal. It is argued that results depend on how one asks, that asking well requires knowing what to ask, and that a plausible answer can lead to under-claiming or letting deadlines expire. The repeated example: the user saves €200 but loses €5,000 by failing to quantify correctly. Some add an uncomfortable warning: there are already convictions against professionals for citing non-existent case law and norms in lawsuits, weakening the argument that error is exclusive to machines.

What the ICAM doesn't say and what the BOE says

The context nobody disputes is the regulatory web. The data cited in the conversation are staggering: Spain publishes an average of 960,000 pages of regulation annually in official gazettes, with about 203,000 from the State and 757,000 from autonomous communities, and over 41,000 state laws between 1970 and 2016. When the legal system is such a labyrinth, demand for someone to translate it does not drop: it rises. And that is where generalist AI finds its niche, with or without Article 403.

The pattern repeating across professions

The discussion quickly shifts to a known pattern. The same technology praised for medical diagnosis, triage, or administrative processing becomes a threat when it touches one's own profession. The overarching question is not whether AI can advise, but who assumes responsibility when it advises poorly. The ICAM has an answer: the licensed attorney. The tech industry has another: the user accepts the terms. Meanwhile, the citizen decides which side to bet on.

The result is that Article 403 could cease to be an empty incivil provision and become the first Spanish rule attempting to incivil cover a product. If it passes, the question will no longer be whether AI replaces lawyers, but whether lawyers can ensure AI still needs them.