The law has always advanced in uneasy tandem with technology, often trailing it, sometimes resisting it, but ultimately reshaping itself in response to what society chooses to deploy in the real world. Few developments test the elasticity of legal reasoning more acutely than the emergence of humanoid systems comprised of things that look and move like a human, even though they are not biologically human. Humanoids are capable of independent learning, adaptive reasoning, and decision-making that increasingly resemble human cognition. This challenge is no longer theoretical. It is already unfolding in airports, in airline operations, and on roads, where artificial agents interact with the public in. spaces historically governed by strict rules of safety, accountability, and trust.
Recent advances in artificial intelligence architecture—illustrated by NVIDIA’s latest generation of chips designed for self-driving vehicles—mark a decisive shift. These systems are no longer mere executors of pre-coded instructions. They are designed to process contextual information, weigh alternatives, anticipate outcomes, and adjust behaviour dynamically in ways explicitly modelled on human cognition. When such systems are embedded in self-driving vehicles, the question of criminal liability is already pressing. When they are installed in airport terminals as check-in agents, security-screening assistants, or—more provocatively—deployed in the future as humanoid cabin crew interacting autonomously with passengers at 35,000 feet, the question becomes unavoidable: if such an entity causes harm through an independent decision, who, if anyone, is criminally responsible?
Is Criminal Law Applicable?
At the heart of this inquiry lies a profound conundrum that criminal law has never had to confront in its present form: the tension between biology and agency. Criminal law has always been anthropocentric. It presumes a human body animated by a human mind, shaped by consciousness, emotion, moral understanding, and socialization. Liability is not imposed merely because harm occurs, but because harm is caused by an agent deemed morally blameworthy. Biology has thus served as a proxy for agency. To be human was to be capable of guilt.
Humanoids disrupt this equation. They are not biological, yet they increasingly exhibit agency. They can perceive, decide, learn from experience, and act in the physical world. When NVIDIA speaks of AI that “thinks like a human,” it is not invoking poetry but architecture: neural networks trained to approximate human cognitive processes, decision trees that evolve, and systems that refine judgment through feedback rather than fixed programming. The law, however, has not yet determined whether agency alone, divorced from biology, is sufficient to ground criminal culpability.
The classical structure of criminal liability rests on two pillars: actus reus and mens rea. The former presents little difficulty in the context of humanoids. A humanoid stationed at an airport kiosk that misdirects passengers, triggers a systems failure, or physically injures a traveller by autonomous movement has clearly committed an act. A humanoid cabin crew member that independently decides to restrain a passenger, administer a sedative, or override a safety protocol would equally satisfy the requirement of conduct. These are not abstract harms; they are physical, observable acts occurring in regulated spaces where safety is paramount.
Mens rea, however, is where the legal edifice begins to tremble. Intention, recklessness, knowledge, and negligence are states of mind as much as legal constructs. They presuppose awareness of wrongdoing, foresight of consequences, and the capacity to choose otherwise. The question is not whether an AI system processes information, but whether it understands normativity—whether it grasps that certain actions are not merely inefficient or suboptimal, but wrongful.
Here, the analogy often drawn with corporate criminal liability is instructive but ultimately incomplete. Corporations, like humanoids, are artificial persons. They lack consciousness, yet they are routinely held criminally liable. The justification lies in attribution: the guilty mind of a human agent is imputed to the corporate entity. The corporation’s liability is derivative. With autonomous humanoids, attribution dissolves. The decision may not be traceable to a particular programmer, manager, or operator. Learning systems may generate outcomes that were neither foreseen nor intended by any human actor. The humanoid’s “decision” is emergent, shaped by data, experience, and probabilistic inference.
This raises an unsettling prospect for criminal law. If no human mind can be identified as the source of the guilty intent, and if the humanoid itself lacks biological consciousness, does culpability evaporate? Or does the law expand its conception of mens rea to encompass artificial cognition?
A Way Forward?
My view, informed by both legal history and regulatory pragmatism, is that criminal law is not yet prepared—conceptually or ethically—to ascribe full criminal culpability to humanoids, even those that think and learn independently. Criminal liability is not merely a mechanism for allocating risk; it is an expression of moral condemnation. Punishment presupposes blame, and blame presupposes a subject capable of understanding censure. A humanoid that recalibrates its algorithms in response to negative feedback may be “corrected,” but it is not morally reproached. There is, at least for now, a categorical difference.
Yet to stop the analysis there would be misunderstanding the trajectory of the law. Criminal law has always adapted when confronted with new forms of agency. The introduction of strict liability offences, particularly in aviation and transport safety, already reflects a willingness to decouple liability from subjective fault in the interest of public protection. Airports and airlines operate within a dense matrix of regulatory obligations precisely because harm in these environments has systemic consequences.
In this context, the deployment of humanoids in terminals and aircraft cabins cannot be approached as a novelty. It must be treated as a safety-critical decision with legal reverberations. Airport managers, in particular, stand at the frontline of this transformation. Their responsibility is not merely operational but normative. They must ask not only whether humanoids improve efficiency, but how accountability will be preserved when something goes wrong.
In the foreseeable future, the sensible legal approach is not to criminalize humanoids themselves, but to embed them within a framework of human responsibility. This means rigorous certification standards for AI systems used in airports, analogous to aircraft type certification. It means clear lines of responsibility linking humanoid actions to accountable human entities—airport authorities, airlines, system manufacturers, and service providers. It also means transparency: decision-making processes must be auditable, explainable, and capable of post-incident reconstruction.
For airport managers, this translates into concrete governance measures. Humanoids deployed at check-in kiosks or information desks should operate under clearly defined scopes of autonomy, with escalation protocols that transfer decision-making to human supervisors when legal or safety thresholds are approached. In security-sensitive areas, humanoids should be treated as decision-support tools rather than final arbiters. In passenger-facing roles, their interactions should be continuously monitored, not out of distrust, but in recognition of the law’s current inability to ascribe moral responsibility to machines.
The prospect of humanoid cabin crew raises even deeper concerns. Aviation law has always emphasized the special status of the aircraft commander and the human judgment required in emergencies. To delegate aspects of this authority to humanoids would require a rethinking of liability regimes under international air law, including the allocation of responsibility for acts committed on board. Until such a rethinking occurs, humanoids in the cabin should be framed as assistants, not autonomous decision-makers.
Nevertheless, the biological versus agency debate will not remain static. If AI systems continue to evolve toward genuine normative understanding—if they can explain their actions in moral terms, anticipate blame, and modify behaviour in response to ethical judgment—the law will face a moment of reckoning. At that point, continued refusal to recognize their culpability may appear less principled than arbitrary. The history of law is replete with moments where personhood expanded in response to social reality.
For now, however, prudence must prevail. Criminal law should resist the temptation to anthropomorphize machines, even as technologists do so rhetorically. Airports and airlines must adopt a posture of heightened responsibility, recognizing that the more autonomy they confer on humanoids, the greater their own exposure becomes. The law’s ultimate concern is not with the ontology of machines, but with the protection of human life, dignity, and trust.
Humanoids may think like humans, but they do not yet belong to the moral community that criminal law presupposes. Until they do, the burden of accountability must remain firmly with those who design, deploy, and profit from them. In the charged environment of airports and aircraft cabins, where safety is both a legal obligation and a moral imperative, that burden cannot be delegated to silicon, however intelligent it may become.
My Take
Having spent much of my professional life examining how law responds to risk in aviation—particularly through the prism of the Tokyo Convention of 1963 and the liability conventions governing surface damage by aircraft—I am struck by how familiar today’s debate on humanoid autonomy feels. We have been here before, albeit in different technological guises. Aviation law has always been compelled to confront situations where harm arises in complex, transnational, and technologically mediated environments, often under conditions where traditional notions of fault and intention are strained. The emergence of independently thinking humanoids in air transport ecosystems is not a rupture with this tradition; it is its logical continuation.
Articles 6 and 8 of the Tokyo Convention remain, to my mind, among the most instructive provisions for legislators grappling with autonomous systems. Article 6 confers upon the aircraft commander the authority to take reasonable measures, including restraint, when he or she has reasonable grounds to believe that a person on board has committed, or is about to commit, an offence jeopardizing safety or good order. Article 8, in turn, empowers the commander to disembark such a person. These provisions are not merely operational rules; they reflect a deeper legal philosophy. They recognize that aviation safety sometimes requires immediate action in the absence of judicial process, and they vest responsibility squarely in a human agent who is expected to exercise judgment, proportionality, and restraint.
The relevance of this framework to humanoids, whether deployed as cabin crew, security assistants, or decision-making agents in airports, is obvious. If a humanoid acts against a passenger, restrains movement, denies boarding, or triggers a security response, it is in effect exercising powers analogous to those contemplated under Articles 6 and 8. Yet the Tokyo Convention presupposes a human commander whose discretion can later be scrutinized against standards of reasonableness. The Convention does not contemplate delegation of this discretion to a non-human agent. This omission is not accidental; it reflects the Convention’s ethical core, which ties extraordinary powers to human accountability.
In my earlier work on surface damage caused by aircraft, I advanced the argument that aviation liability regimes deliberately shift emphasis away from moral blame toward risk allocation. The conventions on surface damage impose liability even in the absence of fault because aviation, by its very nature, creates risks that must be borne by those who benefit from the activity. This logic is instructive for humanoids. Where autonomous systems cause harm—whether on the ground through airport operations or in the air through cabin interactions—the primary concern of the law should not be metaphysical debates about machine intention, but the assurance of compensation, deterrence, and systemic safety.
This is where criminal liability becomes conceptually treacherous. Criminal law, unlike civil liability, is not merely about loss distribution. It is about condemnation. In my book Criminal Liability in Air Transport (Ethics Press, 2025), I emphasized that criminal responsibility in aviation has always been cautiously applied, reserved for egregious departures from accepted standards of conduct. Even then, the law has shown a marked preference for prosecuting natural persons—pilots, maintenance engineers, managers—rather than abstract entities. This preference is grounded in the belief that punishment has meaning only when directed at an agent capable of moral comprehension.
Humanoids, however sophisticated, do not yet meet this threshold. They may simulate judgment, but they do not experience moral doubt. They may optimize outcomes, but they do not wrestle with conscience. To impose criminal liability upon them would risk hollowing out the moral content of criminal law. It would also offer a convenient distraction from the real sources of responsibility: those who design autonomy, define operational parameters, and decide where and how such systems are deployed.
That said, my position is not one of complacency. Legislators cannot afford to wait until humanoids are ubiquitous in terminals and cabins before acting. The lesson of aviation law is that regulation must be anticipatory. The Chicago Convention, the Tokyo Convention, and the liability conventions were all born of foresight as much as experience. They sought to establish principles before chaos dictated outcomes.
The first task for legislators is to resist the temptation to anthropomorphize technology in legal terms. The language of “thinking like a human,” while useful for engineers and marketing departments, is dangerous in law. Legislation should focus on function, autonomy, and risk, not on simulated cognition. A humanoid’s legal relevance lies not in whether it resembles a human, but in the scope of authority it exercises and the magnitude of harm it can cause.
Second, legislators should draw explicitly on the aviation model of layered responsibility. Just as the Tokyo Convention anchors ultimate authority in the aircraft commander, future regulatory frameworks should ensure that decisive powers affecting passenger rights and safety remain traceable to identifiable human decision-makers. Humanoids may advise, assist, and even act within tightly defined boundaries, but they should not become the final locus of discretion in safety-critical matters.
Third, criminal law should be supplemented—not supplanted—by robust regulatory and civil liability regimes. Where humanoids are involved, strict liability models akin to surface damage conventions may be more appropriate than fault-based criminal sanctions. These models incentivize safety by internalizing risk costs without engaging in metaphysical exercises about machine culpability. They also align with aviation’s long-standing emphasis on prevention over punishment.
Fourth, transparency and auditability must become legislative imperatives. One of the greatest dangers posed by learning systems is opacity. Legislators should require that any autonomous system deployed in aviation contexts be capable of post-incident explanation. Not an explanation in a human emotional sense, but technical intelligibility sufficient to reconstruct why a decision was made. Without this, neither criminal investigation nor regulatory oversight can function effectively.
Finally, legislators must address ethics explicitly. Aviation law has always been underpinned by ethical assumptions about human life, dignity, and trust. The Tokyo Convention’s conferral of extraordinary powers is tempered by an expectation of humane judgment. As humanoids enter these spaces, ethical safeguards must be codified, not assumed. This includes limits on force, respect for passenger autonomy, and clear remedies when systems malfunction.
My take is shaped less by fascination with artificial intelligence than by long experience with aviation’s legal architecture. The law does not need to reinvent itself; it needs to remember its own logic. Humanoids should not become scapegoats for human choices, nor should they be elevated prematurely to the status of moral agents. Legislators would do well to approach this challenge as aviation lawyers have always approached risk: by clarifying responsibility, prioritizing safety, and ensuring that accountability remains human at its core. The future of air transport may be increasingly automated, but the ethical burden of that automation must remain firmly grounded in human hands.
