As artificial intelligence (AI) takes center stage in the world’s economic and technological development, a less visible but structurally consequential problem is emerging: the judicial system, as designed in the United States, represents a direct threat to the development of this technology.
The reason lies in the mechanism that regulates civil responsibilities: disproportionate trials, unpredictable verdicts, class actions that multiply the risk and a system of punitive compensation that exceeds any logic of reparation. Faced with this scenario, it is urgent to discuss the possibility of granting partial and limited legal immunity to AI companies in certain aspects, to preserve the ecosystem that makes their existence possible.
The heart of the problem is that AI companies face risks that are not insurable in traditional terms. Insurance works under a principle of dispersion of damages: it is assumed that negative events will be isolated, varied and with more or less predictable costs. But in the case of AI, a single model decision can generate replicated effects across millions of users. An error in model design or training can lead to a flood of identical claims across the country. This systemic and cumulative nature of errors makes it practically impossible to calculate an adequate premium, foresee potential liabilities or establish reasonable coverage limits. That is, insurers cannot—nor do they want—to cover these risks, which leaves companies exposed and vulnerable, with no safety net in the face of a legal storm.
The American judicial system as a structural threat
This is where the American judicial system comes into play. Unlike many Latin American countries, where trials are carried out individually, with written procedures and technical judges, in the United States there are two central figures that drastically raise the level of exposure: the jury trial and the class action.
Jury trial allows twelve citizens with no legal training to decide a company’s liability and the amount of compensation. In complex issues such as those involving AI, this can lead to highly emotional decisions, without technical attachment, inflated by indignation or fear. Added to this is an even more risky figure: collective action.
A class action is a proceeding in which a large group—potentially thousands or millions of people—join together in a single lawsuit, represented by one or more plaintiffs. The argument is that everyone has been affected by the same cause or the same event. In the context of AI, this means that if a model generates a replicable harm – for example, an automated defamation, an erroneous recommendation, or a systematic bias in its responses – all users who have received that same result could join together in a single lawsuit and claim joint compensation. The problem is the accumulation of compensation: what could be a claim of ten thousand dollars is multiplied by millions and becomes a judgment in the billions, impossible for even the largest companies to absorb.
As if this were not enough, the American judicial system allows so-called punitive damages. Unlike traditional compensation, which seeks to repair damage to the victim, these are imposed as punishment to the defendant and as a warning to the rest of society. They are not tied to actual harm, but to the jury’s perception of the seriousness of the conduct. This means that a jury can decide that a company pay ten, twenty or one hundred times the amount of the damage, simply to “teach a lesson.” In an environment where model decisions can be involuntary, statistical, and difficult to predict, the possibility of facing punitive damages turns any lawsuit into an existential threat to the defendant company.
A legal solution already used in other sectors
Against this backdrop, the proposal arises to grant limited legal immunity to AI companies in certain circumstances. This immunity would not eliminate its general liability or allow unpunished abuses, but it would establish clear limits to prevent the judicial system from becoming the executioner of a nascent technology.
Immunity already exists in US law in other areas. For example, foreign governments enjoy jurisdictional immunity: they cannot be sued in US courts except in exceptional circumstances, thanks to a special law. Likewise, the states themselves within the US have sovereign immunity, meaning they cannot be sued without their consent. Public officials also enjoy qualified immunity, which protects them from lawsuits as long as they act within their duties and do not clearly violate established rights.
In the private sector, legal immunities or protections have also been granted in sensitive contexts. For example, pharmaceutical companies that produce vaccines under certain federal government programs are protected from individual lawsuits, and any claims must be brought in special courts, with pre-established caps. Digital service providers also enjoy immunity under certain laws when they act as intermediaries and not as content publishers.
A model of partial and controlled immunity
The model to follow would combine this previous experience: a partial, conditioned immunity, with strict parameters and control mechanisms. Specifically, this exemption from responsibilities would limit three fundamental aspects. First, it would prevent class actions from being filed against an AI company when the damages are diffuse, difficult to quantify, or derived from statistically expected decisions. Second, it would prohibit punitive damages, since the purpose of punishment becomes arbitrary when the harm was not caused intentionally. And third, it would raise the admissibility requirements for lawsuits: not everyone could sue, but rather they would have to present concrete evidence of damage, direct causality, and lack of compliance with technical standards by the company.
A direct and positive consequence of establishing this type of immunity is that it reopens the door to the insurance market. Today, many insurers reject coverage of risks associated with AI because they consider them unlimited, impossible to calculate and economically unfeasible. But if legal immunity is established that sets clear limits on possible compensation, insurers will be able to design specific products, calculate reasonable premiums, establish coverage limits and accompany the development of the sector with financial support instruments. In this way, immunity would not be a shield to avoid liability, but rather a legal framework that allows reactivating the insurance mechanism, essential for the stability of any industry.
This approach does not imply giving companies carte blanche. Quite the opposite. Immunity would only be valid if companies comply with standards of transparency, auditability, security and due care. They could require that their models be reviewed by third parties, that reports be published on their training processes, that consent for the data used be respected, and that there be internal rapid remediation mechanisms for affected users. If these requirements are not met, immunity could be withdrawn. It would be, in practical terms, a conditional immunity, subject to clear rules, as occurs in other areas of law.
The opportunity to act before damage is irreversible
AI will be one of the central infrastructures of the 21st century. As happened in the 19th century with the railway or in the 20th with civil aviation, the legal framework must adapt so as not to kill the technology before it deploys its potential. The United States today has the opportunity—and the responsibility—to create a legal environment in which innovation does not depend on the chance of a jury, the humor of a class-action lawsuit, or the blind calculation of an exemplary punishment.
Partial immunity for certain types of claims can be the difference between a vibrant ecosystem and one paralyzed by fear. It’s not about protecting companies; It is about protecting technology itself as an engine of development and preventing courts from becoming minefields where every invention runs the risk of being destroyed before it is born.
Things as they are
Mookie Tenembaum tackles tech topics like this every week with Claudio Zuchovicki on his podcast Artificial Intelligence, Financial Perspectivesavailable on Spotify, Apple, YouTube and all platforms.
