There is a question that comes up every time a lawyer working in AI governance explains the field to a colleague who has not yet encountered it: if European law and American law are so structurally different in how they regulate AI, why do the arguments look so similar in court? The AI Act and the US Constitution are not comparable instruments. The EU Charter and the Bill of Rights were written in different centuries for different legal traditions. And yet lawyers on both sides of the Atlantic, when they challenge an algorithmic decision affecting a client’s liberty, end up making essentially the same argument.
The answer is that beneath the regulatory differences, both systems are responding to the same underlying problem — and both have developed the same core solution. The problem is that decisions affecting individual liberty require justification. The solution is due process.
In Europe it appears as the right to good administration. In the United States it appears as due process of law. Despite their different doctrinal histories, they serve the same constitutional function: preventing the exercise of arbitrary public power without accountability. When statutes fail to reach a specific harm, when regulatory frameworks contain gaps, when a novel deployment falls outside existing classification systems, due process is the constitutional hook that remains available in both systems. Understanding how it works — and how to use it — is one of the most transferable skills in AI governance practice.
Procedural due process in the United States
U.S. Constitution, Amendments V and XIV (1791 and 1868)
The foundation of procedural due process in the United States lies in two provisions we have already encountered in this book. The Fifth Amendment prohibits the federal government from depriving any person of life, liberty, or property without due process of law. The Fourteenth Amendment applies the same prohibition to the states. Together they create a constitutional baseline below which no government action — however efficient, however technologically sophisticated — can fall.
In practice, procedural due process requires two core elements: notice of the government action affecting the person, and a meaningful opportunity to be heard before an impartial decision-maker. The modern analytical framework for determining exactly what those requirements demand in a specific context comes from Mathews v. Eldridge, 424 U.S. 319 (1976), decided February 24, 1976, with Justice Powell writing for the majority. The Court established a three-factor balancing test: the private interest affected by the government action, the risk of erroneous deprivation under existing procedures and the value of additional safeguards in reducing that risk, and the government’s interest including the administrative burden of additional process.
Mathews is the tool that makes procedural due process analysis in AI cases tractable. Applied to algorithmic decision-making, the three factors produce an analytical framework that courts can use even when the underlying technology is novel. The first factor — the private interest at stake — is typically strong when an algorithm influences detention, deportation, or criminal sentencing. The liberty interest in remaining free from incarceration or removal from a country where one has built a life is among the weightiest interests the Constitution recognizes. The second factor — the risk of erroneous deprivation — is precisely where algorithmic opacity becomes constitutionally significant. When a system is a blackbox whose logic cannot be examined, the risk that an error in its training data or its design has produced an incorrect output cannot be evaluated, explained, or corrected. That ineliminable risk of undetectable error is a powerful argument that additional procedural safeguards — disclosure of the algorithm’s role, access to the factors that influenced the output, human review of the automated recommendation — are constitutionally required. The third factor — government interest — is typically the counterweight. Efficiency and administrative capacity matter. But the argument that automation is more efficient does not, by itself, override the first two factors when the liberty interests at stake are serious.
The Mathews framework connects directly to the immigration AI systems we will examine in Module 5. When ICE’s Risk Classification Assessment recommends detention without meaningful human review, or when the Hurricane Score assigns a risk category that shapes a non-citizen’s ability to seek release, the Mathews analysis is the constitutional structure within which those deployments must be evaluated. Is the person informed that an algorithm influenced the decision? Can they examine the factors that drove the output? Is there human review capable of overriding an erroneous recommendation? The answers, in the documented cases, are often no, no, and nominally yes.
Substantive due process — limits on arbitrary power
Procedural due process asks whether the right process was followed. Substantive due process asks a different question: whether certain government actions are constitutionally impermissible regardless of the procedures used to reach them.
Substantive due process protects core liberty interests — personal freedom, family integrity, the right to remain in the country where one has built a life — from government interference that is arbitrary or that shocks the conscience. It does not require formal procedural failures. It prohibits certain outcomes regardless of how carefully the procedures leading to them were designed.
Algorithmic systems raise substantive due process concerns when they produce outcomes that the government cannot rationally justify in terms that a court can evaluate. If an agency deploys a system that it does not fully understand, that makes recommendations it cannot explain, and that produces outcomes affecting fundamental liberty interests without any possibility of meaningful review, the opacity itself may constitute the constitutional violation. Technology does not create an exemption from the requirement that government actions affecting liberty be rationally explicable and legally accountable. A blackbox that determines who is detained, who is deported, or who receives a harsher sentence is not exempt from substantive due process merely because it processes data faster than a human official could.
The European counterpart — Article 41 and Article 47 of the Charter
Charter of Fundamental Rights of the European Union — OJ C 326, 26.10.2012, p. 391–407
We examined Articles 41 and 47 of the EU Charter in depth in Chapter 7. Here the point is not to repeat that analysis but to identify the structural parallel that makes due process the transatlantic bridge.
Article 41 — the right to good administration — requires that public authorities handle matters impartially and fairly, that individuals be heard before adverse measures are taken, that they have access to their file, and that decisions be reasoned. Article 47 — the right to an effective remedy — guarantees that anyone whose rights have been violated has access to an independent tribunal capable of reviewing the decision. Together, they impose on European public authorities administering AI systems the same three requirements that Mathews analysis produces in the American context: notice that an automated system played a role, explanation sufficient to allow meaningful challenge, and access to human review capable of producing a different outcome.
The language differs. The doctrinal history differs. The institutional architecture differs fundamentally. But the functional requirements converge because both systems are solving the same problem. A decision that cannot be explained cannot be reviewed. A decision that cannot be reviewed is arbitrary. An arbitrary decision by a public authority is constitutionally intolerable in both systems.
The three pillars of algorithmic due process
When procedural due process and the right to good administration are placed side by side, three common principles emerge that apply across both legal systems and that structure the practical arguments available to lawyers challenging algorithmic decisions.
The first is notice. Individuals must be informed when automated systems play a material role in decisions affecting their liberty or status. This is not merely a transparency preference. It is a constitutional prerequisite for any meaningful opportunity to challenge the decision. A person who does not know an algorithm influenced their detention or denial of asylum cannot exercise any of the other rights the legal system provides. In Europe, this follows from the obligation to give reasons under Article 41 and the transparency requirements of the AI Act and GDPR. In the United States, it follows from the notice requirement of procedural due process and the statutory obligation under laws like ECOA — which requires creditors to explain adverse credit decisions — to disclose the basis for algorithmic outcomes.
The second pillar is explanation. Both legal systems require reasoned decisions. The right to explanation under GDPR Articles 13, 14, and 15 — which we examined in Chapter 4 — and the right established by AI Act Article 86 operate alongside Article 47’s requirement that judicial review be meaningful. In the American context, the explanation requirement emerges from the second Mathews factor: if an individual cannot understand the basis for an adverse decision, they cannot contest it effectively, which means the risk of erroneous deprivation is irreducibly high. An algorithmic output is not a reason. It is the beginning of an obligation to provide one.
The third pillar is human oversight. Purely automated decisions affecting fundamental rights create constitutional problems in both systems. We established in Chapter 5 that Article 14 of the EU AI Act requires effective human oversight for high-risk systems — oversight that is genuine rather than nominal, capable of detecting errors and overriding problematic outputs. In the American context, the Mathews balancing test may produce the same requirement: when the risk of erroneous deprivation from an unreviewed automated recommendation is high, and the liberty interest at stake is serious, the additional procedural safeguard of meaningful human review may be constitutionally required.
Rectification as the practical consequence of due process
Once notice and explanation exist, the next question is what happens when the system is wrong. This is where the gap between the European and American frameworks is most visible — and where due process in the American system performs the most work.
European law addresses data errors directly through GDPR Article 16, which provides a right to correct inaccurate personal data feeding an algorithmic system, and through AI Act Article 86, which provides a right to explanation sufficient to identify what drove the output. When a person exercises those rights and discovers that the algorithm relied on an incorrect biometric match, an outdated criminal record, or a misattributed dataset, the path to correction is relatively clear.
American law has no equivalent statutory right at the federal level. But due process provides functional substitutes in specific contexts. In criminal proceedings, Brady v. Maryland, 373 U.S. 83 (1963), requires the government to disclose exculpatory material — and information demonstrating that an algorithmic tool produced an erroneous output that influenced a conviction is exactly the kind of material Brady covers. In immigration proceedings, the Privacy Act of 1974, 5 U.S.C. § 552a, allows individuals to request agency records and seek corrections of demonstrably inaccurate information. Habeas corpus remains available when detention is based on an algorithmic error that no administrative process has been willing to correct.
These mechanisms are narrower and more procedurally demanding than the GDPR’s correction rights. But they exist, they have been used successfully in documented cases, and they connect to the same underlying principle: a system that makes decisions affecting liberty on the basis of incorrect information is making arbitrary decisions — and arbitrary government action is constitutionally intolerable on both sides of the Atlantic.
Translating the argument
The practical value of the due process framework is that it translates algorithmic behavior into constitutional language. Judges do not decide cases based on technical descriptions of model architectures or training data pipelines. They decide cases based on legal principles — fairness, reasoned decision-making, protection against arbitrary state action, access to meaningful review.
Framing algorithmic opacity as a due process problem is not rhetorical. It is legally precise. When a system produces an outcome that cannot be explained, when the individual affected has no meaningful opportunity to contest the basis for the decision, when human review is nominal rather than genuine — those are due process failures, cognizable under both the Fifth and Fourteenth Amendments in American courts and under Articles 41 and 47 of the EU Charter in European proceedings.
The technology is new. The constitutional problem is not.
Next: Chapter 13 — The right to explanation. Your primary weapon against the blackbox.
Leave a Reply