Introduction

How often do you think about Roman Law? To all of us who have spent most of our graduate life dedicated to the study of law, and thrived in it, we understand better than anyone else when someone asks How often do you think about the Roman Empire? In truth, for the legal scholar, the question is less about how often and more about when we do not.

The very foundations of modern legal thought can be traced to the Romans (Samuel, 2014). Indeed, in the year 2025, amidst the fleeting pronouncements of contemporary legislatures and the clamour of novel legal disputes, the enduring wisdom of Roman jurisprudence stands with quiet, almost insolent authority, demonstrating its continued utility for future legal designs, particularly concerning Artificial Intelligence (Deibel, 2021).

It is within this venerable bedrock upon which our intricate legal architecture continues to rise, a testament to its singular genius for forging law as an intellectual artefact, a matter of knowledge, not mere ephemeral experience, designed not merely to regulate, but to secure order and banish chaos through a system of calculable reason (Cinelli, 2021).

The Roman jurists themselves were involved in systematising knowledge and reducing it to easily understood propositions, a tradition that ultimately led to the codes of Continental Europe (Keep this in mind, since the next chapter of the series will explore the European Union and AI) and influenced legal scholarship in multiple ways. The most recent statement should come as no surprise; we are all quite familiar with the idea by now that legal systems have often taken laws or ideas from other legal sources.

For instance, Solon's laws for Athens in the sixth century B.C. were influenced by the legal codes of other Greek city-states. Much of the English Statute of Frauds of 1677 was modelled on the French Ordonnance de Moulins. The 1960 Civil Code of South Korea drew heavily from German law. Perhaps the most significant borrowing occurred when European lawyers introduced into medieval society the ancient Roman law contained in Justinian's Corpus Juris Civilis (Mather, 2002). This article aims to draw from that vast fountain of Roman Law, those concepts that, by some means or another, have survived the passage of time, and can help us better understand the artificial development of the future. So, buckle up, and let’s head back to the future.

​

The Accountability Gap and the Relevancy of Roman Jurisprudence

​

A central issue posed by AI is the "accountability gap" or "responsibility gap". This problem arises when AI entities, capable of autonomous decision-making, cause harm in any way possible. Here, traditional legal frameworks have struggled to assign responsibility to a human or corporate actor. The increasing autonomy of AI systems, from the well known chatbots to AI Romantic Companions (one can notice that while the former were already well known, the latest is getting more and more popular (Willoughby & Carroll, 2025), and this scares the writer quite a lot), necessitates urgent legal discourse on who bears the risk of erroneous declarations or actions (Maatz, 2024). While many AI systems still have human oversight, the growing number of fully autonomous systems demands legal clarity for both the users and developers.​​

​

Remarkably (and fortunately for us), this challenge is not entirely novel (Deibel, 2021). Legal scholars have identified striking parallels between the dilemmas posed by AI and those faced by ancient Roman law in regulating slave-run businesses (Cinelli, 2024).

Before plunging into the dry labyrinth of Roman law as it pertained to slaves, let us first pause for a moment’s panoramic view. Allow me, in a few paragraphs, to sketch the broader picture upon which the drama of servitude unfolded. I aim to offer the uninitiated reader, perhaps unaccustomed to the peculiar institution of ancient slavery, a glimpse into the Roman mind. In short, I shall attempt to restore some measure of humanity to a class of people whom history has all too often reduced to mere shadows, perhaps the most dehumanised population ever to walk the earth. Among the numerous scholars and writers who have addressed the complexities of Roman slavery, I have found the prose of Edward Gibbon, author of the seminal History of the Decline and Fall of the Roman Empire (1776), to be particularly distinguished for its clarity, analytical precision, and subtle but unavoidable wit. Accordingly, it is Gibbon’s perspective that will provide the principal framework for our exploration of this topic.

We have to take into account that the majority of slaves were barbarian captives, taken in great numbers by the fortunes of war, “purchased at a vile price, accustomed to a life of independence, and impatient to break and to revenge their fetters.” This influx of enslaved peoples contributed to a population dynamic where, as Gibbon notes, “the slaves were at least equal in number to the free inhabitants of the Roman world.” When the Emperor Claudius (now better known for the Robert Graves novel I, Claudius) did a census it recorded nearly seven million Roman citizens, and when accounting for women and children, the total free population approached twenty million, yet the number of slaves remained uncertain and fluctuating, underscoring their immense but neglected presence.

I find the concentration of slaves within places simply astonishing. We have to take into account that “on a very melancholy occasion, four hundred slaves were maintained in a single palace of Rome,” a testament to the reliance of elite households on vast enslaved labour forces. Even freedmen, once emancipated, could amass considerable wealth and retain large numbers of slaves, as exemplified by one freedman under the reign of Augustus who, despite losses in civil wars, left behind thousands of cattle and “four thousand one hundred and sixteen slaves,”. The conditions endured by slaves were often harsh and brutal, but some of them, mostly Greek in the Roman world, were used as teachers for the children, so their living conditions were considerably better. In the free states of antiquity, “domestic slaves were exposed to the wanton rigour of despotism,” and “the most severe regulations and the most cruel treatment seemed almost justified by the great law of self-preservation.” This the harshness reflected the widespread fear among masters of rebellion and unrest, justifying strict control measures. However, over time, the legal and social handling of slaves evolved. The value placed on a slave’s life increased, and although their well-being still largely depended on their master's disposition, “the humanity of the latter, instead of being restrained by fear, was encouraged by the sense of his own interest.” Furthermore, the ultimate authority over a slave’s life and death, “a power long exercised and often abused,” was gradually taken from private hands and “reserved to the magistrates alone,” signalling a shift towards greater regulation and some degree of protection within the Roman legal framework (Gibbons, 1776).

Now let us get back on our legal track.

Roman jurists grappled with how to manage slaves who, despite being legally classified as res (things or property), functioned as rational, autonomous economic agents (Heine & Quintavalla, 2024). This dualism of persona (person) and res, the simultaneous status of being human and a thing, or a person and property, provided a pragmatic framework for navigating complex issues of agency and liability (Deibel, 2021).

The core insight from Roman law lies in its development of a "heterogeneous framework" of pragmatic legal remedies called actiones adiecticiae qualitatis to bridge this accountability gap. These constitute a set of legal remedies established by the praetor to enforce the liability of the pater familias for obligations incurred by either his son or his slave in various circumstances.

While our primary focus here is on those actions regarding slaves, it should be noted that these remedies were not exclusively limited to them. Roman law offered a diverse "menu of legal options" that allowed for context-specific regulatory responses, reflecting a continuous process of risk allocation among stakeholders (Heine & Quintavalla, 2024). We are going to explore them in the following paragraphs.

 

Actiones adiecticiae qualitatis

One of the most instructive Roman institutions is the peculium, which is a separate fund or asset pool managed by a slave but ultimately owned by the master. The peculium served as a cap on the master’s liability for the slave’s transactions, limiting risk while enabling economic activity (Heine & Quintavalla, 2024). This concept has clear resonance for AI governance. The idea of a “digital peculium” suggests the establishment of fixed capital requirements or liability limits for AI applications, thereby fostering innovation without exposing owners to unlimited risk (Deibel, 2021). In a similar style, the Roman actions known as actio exercitoria and actio institoria made masters liable for business obligations entered into by slaves acting as managers, but only within the scope of explicit authorisation (Heine & Quintavalla, 2024). This approach underscores that AI liability should be context-specific, making owners responsible only for actions they have explicitly authorised. The actio quod iussu further refined this framework by imposing full liability on the master when a slave acted under a specific order, directly linking responsibility to explicit human authorisation (Heine & Quintavalla, 2024).

While this principle is attractive for AI regulation, it also raises concerns about creating overly restrictive environments, so-called “walled gardens”, that could stifle the efficient use of AI and exacerbate the unpredictability associated with complex, opaque systems (Heine & Quintavalla, 2024) Another relevant Roman remedy, the actio de in rem verso, imposed liability on the master if he benefited from a slave’s contract, even absent prior knowledge. This doctrine offers a powerful analogy for modern “association risk” and “network risk” in AI, where owners may profit from AI actions in ways that are difficult to trace or anticipate, such as in cases of algorithmic collusion or collaborative AI networks. By holding owners liable for benefits received, Roman law incentivised oversight and responsible management, a principle that remains relevant for AI governance.

Finally, the actio tributoria addressed situations where a slave operated a business with the master’s knowledge and the master appropriated part of the peculium, requiring the master to share liability with creditors according to court-determined quotas. This remedy provides a useful model for distributing liability in cases of AI “network risk,” where multiple parties may benefit from or contribute to the actions of interconnected AI systems. These Roman legal analogies are not intended as direct solutions or as arguments for reviving ancient institutions in their original form. Rather, they demonstrate the enduring utility of careful legal analysis and policy design in managing the risks and responsibilities that arise from delegating agency, whether to slaves in antiquity or AI systems today.

​

The Modern Debate: AI as Persona or Res?

​

Today's legal scholarship grapples with the fundamental question of AI's legal "status", are AI systems sophisticated objects (res), or should they be treated as legal persons (persona), similar to humans or corporations, or perhaps a new, sui generis category (Mocanu, 2022; Powell, 2021) Currently, AI is largely considered res, with liability falling on its producer, operator, or owner, who are natural or legal persons. However, the "personification" of computers and robots is already a topic of discussion (Maatz, 2024). Arguments for granting AI legal personhood stem from its growing autonomy, decision-making capacity, and the need to address the accountability gap (Haim, 2019). Proponents suggest that recognition could be "direct" (based on moral recognition of autonomy) or "indirect" (treating algorithmic entities like traditional organisations). The "bundle theory" of legal personhood offers an adaptation that could accommodate AI by analysing its status in terms of "ultimate value," "liability," and "commercial" contexts (Mocanu, 2022). This theory, along with the concept of "partial legal capacity" or Teilrechtsfähigkeit, acknowledges that legal personhood might be a "cluster concept" with gradient possibilities rather than a binary state (Mocanu, 2022). Granting AI legal recognition, even as a director or consultant in corporate governance, is seen as having immediate pragmatic benefits (Powell, 2021).

Notwithstanding this, many scholars resist granting AI full legal personhood. Critics, including experts in robotics, AI, law, medical science, and ethics, warned the European Commission against granting robots rights, arguing such proposals seemed influenced by science fiction rather than real-world experience (Haim, 2019) here one just wonder how difficult the work of a lawyer would be in the world of the blade runners!. One key objection is that AI, unlike Roman slaves, lacks sentience, c'est-à-dire the capacity to experience feelings and sensations (Deibel, 2021). There’s also the hard Problem of consciousness, explaining why certain brain states give rise to experience, but this remains unsolved, complicating any definitive claim about AI sentience (Hills, 2020). Without consciousness, AI's "preferential behaviour" might just be sophisticated mimicry of human responses, akin to a robot vacuum cleaner announcing a low battery or complaining when the opposite team scores a goal. The European Parliament, for instance, has moved away from advocating for "electronic persons", stating "no need" currently exists (Mocanu, 2022), but this will be explored in further articles.

Instead, the analogy of AI as "agents" or "servants" of human principals is often favoured for liability purposes (Haim, 2019; Nanos, 2020). This approach emphasises AI's instrumental role and its service to human purposes. It allows for the application of strict liability regimes, such as respondeat superior, making the human principal responsible for the AI agent's actions, and ensuring victims receive remedies. Some argue that manufacturer liability, a form of strict liability, is aligned with the Roman concept of potestas, as producers are best positioned to take preventative measures (Wagner, 2019).

​

The Path Forward: Flexibility and Historical Guidance

 

Ultimately, law, as an artifactual system, must continuously adapt to the evolving realities of technology (Mocanu, 2022; Siliquini-Cinelli, 2024). The Roman experience teaches us that a single, monolithic solution (here, the writer is not referring to the giant and daunting obscure monolith that appears in Kubrick’s 2001: A Space Odyssey) for AI liability is unlikely to be effective. Instead, a "heterogeneous framework" with multiple remedies, tailored to specific contexts and evolving needs, is preferable. This dynamic approach allows for regulatory experimentation, fostering innovation while continually balancing the interests of business and society. As AI continues to advance, the legal field must remain a flexible and scholarly pursuit, drawing on its rich history to illuminate the complex path ahead (Cinelli, 2024) using the light of the past.​