We Keep Debating AI Law and AI Governance Separately, but That’s the Mistake

Every few months, headlines repeat the familiar debate: a self-driving car crashes with no clear responsible party, a hiring algorithm subtly discriminates, or an artificial intelligence (AI) misdiagnoses cancer that a junior resident would detect. Responses typically follow established patterns: lawyers focus on who can be held liable, technologists discuss system design and ethicists debate the moral implications. These conversations seldom intersect, which is exactly why they fail to produce solutions. In fact, AI law should be an integral part of AI governance.

Scholars emphasize an important but often overlooked point: debates about AI and the law, AI governance and the balancing problem are not isolated discussions. They form three interconnected layers of the same issue. Treating them as separate fields is a main reason why many AI policies fail.

The three layers of responsible AI

The law establishes minimum standards by defining liability, safeguarding rights, and ensuring due process in the event of issues.

Governance guides the overall system by embedding core values, setting data procedures, creating oversight structures and proactively identifying potential risks to prevent harm.

Governance should ensure that explanations are trustworthy and accurate, and laws must provide an effective way to challenge decisions. Neither governance nor law can replace the other’s role.

Balance disciplines in decision-making by acknowledging the unavoidable tensions between speed and safety, transparency and security, innovation and caution, as well as efficiency and fairness. Ignoring these trade-offs constitutes a policy failure. Collectively, these three layers form a coherent AI governance framework.

Where the disconnect becomes visible

Consider autonomous vehicles or weapons. When issues arise, existing negligence or international humanitarian law often cannot clearly assign responsibility among manufacturers, developers, operators and users. While this seems like a legal challenge, it stems from governance issues. Technical standards develop quickly, but legislation lags behind. Courts end up facing problems that governance has not foreseen.

Predictive policing and AI-assisted sentencing both show similar issues. An explainable algorithm can still produce misleading results, while full transparency might make systems vulnerable to manipulation. A defendant’s right to a fair hearing is not fulfilled simply because an algorithm can explain itself. Governance should ensure that explanations are trustworthy and accurate, and laws must provide an effective way to challenge decisions. Neither governance nor law can replace the other’s role.

AI challenges extend across every sector

This pattern appears across nearly all areas influenced by AI. Assessing whether AI-generated content counts as original for copyright involves not only legal considerations but also governance questions about whether the system genuinely creates new outputs or merely reuses patterns from its training data.

Whether biased medical AI leads to malpractice liability depends on legal standards and whether governance considers the bias a preventable design flaw or a sign of deeper structural inequalities in the data.

Environmental regulation faces this same challenge. Policymakers struggle to regulate AI’s carbon footprint effectively unless engineering decisions, such as choosing central processing units or graphic processing units, model designs or training methods, are recognized as policy-relevant, not merely technical details.

Why law alone is never enough

This uncovers a deeper truth: law typically comes into play after harm happens. It assigns responsibility but seldom prevents harm in advance. Governance works earlier by influencing design, incentives and oversight. However, without legal authority, governance often cannot be enforced.

Numerous ethical frameworks effectively list risks and principles but often do not specify accountability for harm.

Balance is equally crucial, requiring policymakers to acknowledge unavoidable trade-offs. Merely recognizing a trade-off is not enough to resolve it. The three components function collectively: Balance emphasizes the trade-off; Governance establishes institutions and systems to manage it; and Law provides legitimacy, accountability, and enforcement. Each layer offsets the weaknesses of the others.

The cost of treating them separately

Sadly, many current AI discussions assume that a single layer can substitute for others. Industry often advocates for self-regulation because it is technically knowledgeable and adaptable. Although quick responses are beneficial, self-regulation lacking democratic backing and enforceable accountability is not true governance; it simply reflects preferences.

Governments frequently create laws without clearly outlining the trade-offs involved. As a result, such regulations seem thorough but can face difficulties when applied to complex real-world scenarios.

Meanwhile, numerous ethical frameworks effectively list risks and principles but often do not specify accountability for harm. As a result, Responsible AI remains more an ideal than a measurable, enforceable system.

Designing AI governance in the right sequence

The implications go beyond mere theory. AI-driven legal aid will fail to reduce access-to-justice disparities if governance is controlled solely by the companies creating these systems. Labor protections depend on governance, the pace of automation and the establishment of safeguards. A welfare applicant’s ability to contest an automated decision is only effective if the system was built with explainability and auditability in mind from the beginning, not as an afterthought.

Effective governance depends on the integration of law, governance, and trade-offs. Without law, governance cannot be enforced, and without any of these three, we merely recognize tough decisions. 

The lesson is not about paralysis; it is about sequencing. Before legislators draft the next AI law, policymakers should first determine the trade-offs that the law aims to address. Similarly, before engineers establish governance frameworks, they need to clarify which legal rights those frameworks should ultimately uphold. Both groups must ensure that, before declaring success, the public can still pinpoint who remains responsible when AI causes harm.

One integrated framework

AI does not force us to choose between law, governance, and balancing values; instead, we should see these as interconnected parts of a single framework. Without governance, law tends to be reactive.

Effective governance depends on the integration of law, governance, and trade-offs. Without law, governance cannot be enforced, and without any of these three, we merely recognize tough decisions. When they function together, AI can stay innovative, trustworthy and accountable. AI does not compel us to select between law, governance and difficult choices; it reveals that such a choice was never realistic.