AI Regulation: A Guardrail for Human Control

Artificial intelligence (AI) is here. Systems that once produced simple text can now write software, analyze enormous quantities of information, conduct research, imitate human communication, and perform tasks often without human scrutiny. Businesses, government agencies, courts, and other institutions are already using AI, often in ways that are not readily apparent to the people using it. Most people also interact with some form of AI every day through internet searches, recommendation systems, virtual assistants, and other digital services. These tools offer substantial benefits, but they also raise a fundamental question: Who controls the decisions that shape their design, deployment, and use?

That question should be at the center of AI regulation. AI systems do not choose their purposes, decide where they will be deployed, or determine how much authority they will receive. People and institutions make those choices. Descriptions of an AI model “escaping,” “deceiving,” or “resisting control” can wrongly suggest that it possesses human intentions or an independent judgement. It doesn’t, it can only produce results based on its programming, training, objectives, available information, and access to other systems. Those results may sometimes be unexpected, or difficult to predict, but they always arise from human decisions about how the AI platform was designed, trained, deployed, and controlled. Regulations are needed to ensure that those decisions remain visible, reviewable, and accountable and that responsibility remains with the people and institutions that created or authorized the conditions in which the system operates. JSI examines these questions further in its Justice & AI Resources.

The Risks of Giving AI Too Much Authority

The risks associated with AI range from harms that are already occurring, including fraud, deepfakes, privacy violations, and discriminatory automated decisions, to more speculative concerns about what increasingly powerful systems may eventually be capable of doing. One highly publicized concern is that a future AI system might cause catastrophic harm. That possibility seems doubtful given the current state of AI, and there is a serious question about whether AI will ever develop the independence or capabilities necessary to threaten humanity. 

Even that concern ultimately returns to human control. A system does not need to become conscious or malicious to cause serious harm. People can give it a poorly defined objective, connect it to sensitive information or critical infrastructure, allow it to act without meaningful review, or use it deliberately for destructive purposes. The immediate danger is therefore less about a machine independently deciding to take control than about people surrendering too much authority to systems they do not fully understand or using those systems irresponsibly.

Transparency Is Essential to Human Control

The primary case for regulation does not depend on predictions of global catastrophe, but on the reality that AI already influences the information people receive, the choices presented to them, and decisions involving employment, credit, healthcare, education, public benefits, and interactions with government, often without their knowledge. When individuals do not know that an AI system is involved, what information it uses, how it reaches a conclusion, or who is responsible for correcting an error, human control has little meaning. Meaningful human control requires the use of AI to be disclosed and its decisions to be subject to explanation, review, and reconsideration. 

Human Responsibility in Courts and the Justice System

These concerns are especially serious in courts and in the broader justice system. An AI system may help research legal issues, evaluate evidence, assess risk, manage cases, or recommend decisions. Each use begins with human choices about the system’s data, purpose, limits, and role. No court should treat an AI output as neutral simply because it was produced by a computer. Judges and other justice professionals must remain responsible for determining whether a system is reliable, whether its use is appropriate, and whether its recommendation is consistent with the law and the facts of the individual case. When liberty, public safety, and equal treatment are at stake, a human decision-maker must retain the authority and ability to question, reject, and explain the result.

A Risk-Based Model for AI Regulation

The European Union’s (EU) AI Act provides a useful model for preserving human control through risk-based regulation. It does not impose identical rules on every application. Most minimal-risk systems remain largely unrestricted, while certain systems must disclose that a person is interacting with AI or viewing AI-generated material. It prohibits specified practices, including social scoring, harmful manipulation, and certain uses of biometric identification. These rules recognize that people cannot exercise meaningful control over technology they cannot identify or understand.

The EU  imposes its strongest requirements to AI uses that can significantly affect safety or fundamental rights, including applications involving employment, education, essential services, law enforcement, critical infrastructure, and the administration of justice. High-risk systems face requirements concerning risk assessment, data quality, documentation, activity logs, accuracy, cybersecurity, and human oversight. The purpose of these safeguards is not merely to improve technical performance. It is to give people enough information and authority to supervise the system, intervene when necessary, and determine who is responsible when something goes wrong.

The EU framework also recognizes that a simple spam filter should not face the same controls as a system that evaluates evidence, recommends a sentence, or helps determine eligibility for a public benefit. This model allows ordinary, low-risk tools to remain useful while directing stronger safeguards toward systems that can materially affect a person’s rights, opportunities, or freedom. The greater the authority delegated to AI, the stronger the requirements for human review and accountability should be.

Oversight Must Continue After Deployment

It is important that this oversight must be ongoing. An AI system’s performance may change as its data, users, and operating conditions change. Human control therefore requires continuing monitoring, periodic testing, incident reporting, and the ability to suspend or withdraw a system that no longer performs as intended. Approval at the time of purchase or deployment is not enough.

Artificial intelligence can improve productivity, expand access to information, and help institutions serve people more effectively. Regulation should preserve those benefits while setting clear limits on the authority delegated to machines. It should also prevent organizations that ignore safety and accountability from gaining an advantage over those that invest in responsible practices.

The Central Question: Will People Remain in Control?

The important question facing humanity is not whether AI will someday take control. It is whether people will retain control over the systems already influencing human lives. That requires rules establishing when AI may be used, what decisions must remain with people, what information must be disclosed, and who is accountable for the result. Meaningful AI regulation should ensure that technology supports human judgment without replacing human responsibility.