Artificial Intelligence, like many other pseudoscientific terms, does not have a single interpretation or a strictly defined definition. In this research, the term 'Artificial Intelligence' will be considered in the sense of a sub-field of computer science that seeks to understand and develop systems that show intelligence. Problem-solving systems or speech recognition systems, for example, are considered 'intelligent' systems, which is why they have often been classified as belonging to the field of AI. In the literature, there are different ways to classify AI according to different categories of Artificial Intelligence:

The main categories consider the following types of AI, these being the forms of classification that have been proposed by at least two or three authors:

We will briefly discuss each of these categories, pointing out the distinctive features of the systems that belong to these categories. There is a significant difference between these four types of AI. While Reactive Machines are capable of making decisions, they do not store data from the previous ones. The systems from Limited Memory and Theory of Mind are based on the analysis and accumulation of information on certain interpretable contexts. Finally, Self-awareness systems are the most sophisticated because of self-reflection. The difference should be significant concerning the liability law and the obligations of self-aware systems or platforms as machine guards (Dajeh, 2024a).

Over the last few years, important research has been done about how AI can be used to improve safety systems in a smart factory. This work illustrates the potential of AI used in machine guards. One of the most important applications can be the prediction of maintenance. This is very useful in the machine guard because the safety system may need maintenance more often than other devices. When maintenance is not predicted in advance, the safety system can go down, provoking a machine stop, which results in operational loss for the factory. Likewise, this use case can also be applied to protect the safety system itself, which usually protects the operators. AI can be used, for example, to detect physical anomalies in some parts of a light curtain before a failure.

Furthermore, machine learning can be used to alert the system that an application was not planned, and there would be a risk. In general, event anomaly detection is very important for machine guards. Whether in manufacturing use cases or not, there are general use cases where AI algorithms are used in safety applications to either ensure a safer environment for the workers by ensuring a fast shutdown of a machine or ensuring faster fault detection. All these use cases have in common that they impact end-user safety but are also critical for efficient operation by avoiding the degradation of productivity. More generally, most of the research presently done about AI in safety focuses on integrating AI into conventional security systems. This trend has the advantage of providing a quick response to security threats, but it also has negative impacts since it increases the attribution of responsibility. Indeed, as intelligence increases, demonstration takes more and more precedence in decision-making. According to law, manufacturers are liable if their products are defective and this has caused personal injury or injury to items of use. Thus, if the machine enters a state of disrepair and if the decision process implemented is shown to be defective, the question that arises is whether the use of AI would increase the demonstrability requirement (Dajeh, 2024b).

In the construction of a machine or system for controlling it, its designer or manufacturer must comply with certain legal obligations, such as safety, traceability, information to be collected, and technical documentation. Sometimes this obligation is also extended to the user after good technology and compliance with the instructions for use. The starting point for the machine guard from a civil liability position is the concept of duty of care. This principle sets out the obligations to behave in a manner consistent with the behavior of other players in human society to trigger legal responsibility.

The normative standard connected to the concept of duty of care is given by the imperium of law made of an outline of the latter day of technological developments that could make useful a deep innovation in the type of behavior held by design and assembly firms. The thing guarded, in this case, is the machine equipped with an artificially intelligent element. Therefore, one could consider that the standard is always that of the obligation to produce a machine guardian number or, at most, that regulates the obligation to produce a machine guardian according to the principle of a thinking being. In the light of these, it must be admitted the importance of the supervisory and controlling plan in the liability of the company that produces a machine guardian with an attitude and possibly independent thought. A new reinforcing effect would possibly be attributed to regulatory bodies on the concept of duty of care to be held by those who design or make. Lastly, the necessity of a list of definitions to be put in place by a legislator is pointed out in order to be able to mix current regulations and future possibilities created by AI technologies in the legal field, to be certain in ways that robot line entrust limits to those who choose to adopt AI for making responsible and responsive robots. These definitions in a normative scheme are important because ambiguity in the liability of some logic of never-ending would be pernicious.

The shift to AI-driven machine guards presents various issues under current liability principles, such as the issue of there being no human intervention. In traditional systems, lack of human intervention means that the machine guard is not in use. As a result, the traditional guard would not be operational. When malfunctions happen, it stands to reason to hold the employer liable for failing to keep the guard in place. In AI-driven machine guards, however, the lack of human intervention implies that the human 'intervention' simply did not take place. It is not a malfunction of a guard that can be held against the employer. It is, in fact, the independent malfunction of the AI. In these cases, pinpointing liability becomes more complicated.

The AI's autonomy also clashes with two traditional liability principles. Foreseeability of AI malfunction is problematic, as the unpredictability of an AI system increases with the autonomy of a self-learning system. Hence, it becomes more difficult for a developer to minimize the risk of damage, as he can no longer foresee what his system can do. In principle, control could be exercised by a user over the AI. And temporary control will and can still exist. However, in the long run, AI that is fully capable of taking on all tasks related to a machine guard requires no human input to function. Hence, this offers a way for the manufacturer/developer to avoid liability through navigating the use of these traditional principles.

As AI is increasingly capable of independent operation, liability rules built on the necessity of preventative human oversight might refashion AI manufacturers to bear the risk of harm generated by autonomous systems, and the 'human safety net' may degrade to a somewhat precarious guarantee. It is the policy of requiring a safety-interventionist human that no longer aligns with our contemporary robot lawmaking. Hence, AI cannot be integrated under the same legal regulations used in the traditional machine guard environment.

Legal Surroundings.

The United States does not have specific legislation and regulations in place that would regulate the use of AI in machine guards. However, a series of trade standards in the relevant regulatory framework establish safety rules and refer to liability. Injury and litigation are also governed by general provisions of the civil code, including rules applicable in tort. The United States has a comprehensive and detailed civil liability system. In addition to the legal standards in the code, judicial opinions, known as precedents, are also relevant.

The US legal system is based on a common law system, which means that courts are important sources of law. The way in which liability law is to be interpreted and applied varies from state to state in the United States. In many cases, several competent judges make decisions, and state courts in the first instance may refer to the opinion. This makes it extremely complex to predict the direction of the case law and interpret judgments. Principles, changes, and developments in the law are also made part of the law through the opinions of legal authors. This state of affairs examines the use of AI as a measure of the general legal system in civil liability, tort liability, and other liabilities in order to identify certain features of the legal system required to address issues related to the use of AI in a machine guard. There is also an overview of the ongoing discussion on the establishment of civil liability in new technologies, which can lead to the correction of certain principles of liability. Finally, it examines the extent to which legal certainty can be established through a discussion of the relevant questions related to AI and the application of general principles to the judicial process (Welser, 2017).

The US legal system is essentially rooted in the philosophy of protection of individual autonomy; actors are generally responsible for the harmful consequences of their actions. The US introduced this system of strict liability with a notable case. This strict liability is now organic because it is detailed in a specific section of the Restatement of Torts. This instrument fixes a two-class system, the burdens. Actually, it creates a prima facie case, and then the right to adduce the evidence is shifted to the defendant. This notion of liability without fault also holds for manufacturers submitting to the customs of a standard safer community.

The relevant section of the Bürgerliche Gesetzbuch has been created to be flexible, and the jurisprudence fills out the broader details. The German legal philosophy articulates the legal system in providing broad legal resiliency to cover risk liability, and in particular, to control industrial risk, but also the liability for owners and possessors of anything. According to the Bürgerliche Gesetzbuch, the liability is due even when there is no fault in the sense of its esteemed theater facts. The absence of fault is compensated for by a legal fiction that guilt is established when the victim is injured. The liability of manufacturers is almost strict by analogy to another section of the BGB. This regulation is mitigated by the concept of built-in oil. The manufacturer has indeed the right to prove that the defect comes from a lack of scientific reconnaissance and was therefore unknowable at the time when the product was placed on the market. This liability is almost strict, as first of all, the consumer doesn’t need to prove the fault of the repairer; the absence of any fault by the injured has been compensated by the introduction of infallible absolute proof when the infringement was suffered (Towards [...], 2024).

In the area of emerging technologies, due to the supranational character of the European Union and the increasing technological convergence in the internal market, considerations of harmonization are constantly in the background of the discussion on the specifics of the rules of machine guards. This subsection analyzes the settings defined for the operation of a machine guard, discussing the layer of both civil liability for the damages inflicted by the guard and for the obligation of the machines with respect to their user protection. The latter is based on the most comprehensive regulatory framework defined on a global level as of the date of preparation of the present paper, which is aimed at covering aspects inherent to the new technology. In addition to the general guidelines, there is no shortage of sectoral directives, soft law tools, and case law which are interconnected, hence contributing to the delineation of the relationships between the subjects involved. The most recent proposals provide for the creation of a regulatory framework specifically designed for AI and robotics and, more generally, of a European agency to support and supervise the single national authority in the verification of the conformity of the AI system with European legislation in the field. The approach of the European Union can be compared with that regarding the guards. Both are based on the principle of safety by design, the distinction between low- and high-risk systems, and the composition of a mandatory code of conduct. Also, while waiting for the introduction of a regulation on vulnerability in the EU, some legislative interventions made by Member States require the introduction of features that, in broad terms, are found in the machinery directive (such as transparency, ethics, traceability, explanation, and minimization criteria). Differences exist especially at a general level, due to the different reference to the extent of respect for fundamental rights, ethical standards, and jurisdictional assessment. According to some of these legislative texts, the courts could carry out an evidential assessment of the AI system in the light of the technologies applied to it, in the manner of traditional expert advice. Other legislative provisions at a national level do not involve any legal consequences but leave room for the introduction of self-regulatory codes of a professional or corporate nature, with a view to ensuring compliance with ethical standards. These differences appear to be the result of the complexity of the task of aligning the laws of the twenty-seven jurisdictions, as well as belonging to non-homogeneous legal systems, and are based on diverse traditions which, according to some, represent the basis of the differentiated approach characterizing the European Union’s political strategy on artificial intelligence, which seeks to create mutually reinforcing mechanisms with its Member States. Thus, the proposal of a horizontal AI Act at the level of the twenty-seven Member States would be widened with EU added value in several respects, setting the floor under a vision where the interest in compliance with fundamental rights and high ethical standards is more explicit and, at the same time, technology-neutral. In accordance with the spirit of the machinery directive, it is actually the type of oversight that matters, hence the specific function of a machine guard. More specifically, the machine guard can provide evidence of a malfunction, breach of the obligations inherent in the guard’s own function, and/or guilt of the injured party. Being, however, connected to the control functions of the machine, a duty of the guard to monitor the practical impact of the relative rules – except for sectoral regulations – cannot be ruled out (Wendehorst, 2020; EU, 2024).

In principle, the scope for legal innovation could also be justified under civil liability laws. The principle of strict liability asserts that liability lies not in the fault of an actor or agent—then accepted as a way to limit damages ex post—from a principled point of view but instead in legal space for considering ex ante how funds might substitute liability for or mitigation of precaution. The concept of strict liability has not been explicitly formulated within extant laws as a potential recourse to harm caused by AI-driven machine guards. However, the doctrinal concept of strict liability accords with judgments on, for example, guarantees and omissions of consent within the civil and criminal legal fields where machine activity is not restricted to the AI context of this paper but might be instructive (Wendehorst, 2020).

In this manner, a strict liability regime would refocus the legal lens from the debate about fault and impermissible risk to the causation and certainty of the damage, thus upending foundational components of the legal calculus about negligence, reasonable precautions, and risks of AI. While strict liability does not solve problems of causation, its adoption would simplify the question about verification of the claimant’s damage, and at times the extent. Although such a scheme is more direct, a persistent challenge with it is the articulation of certain types of damage and their instantiation with existing mode data. However, one of the potentials of this regime is that when it is obvious that the claimant has suffered a harmonized damage that AI has clearly caused, linked to matters such as perception in decision making, doubt about the existence of damage is remarkably reduced (Godi, 2024).

Google car accidents. Few of these self-driving vehicle accidents make major news as, although they occur with regularity, they almost always result in minor fender benders. However, two accidents do qualify, in that both resulted in minor injuries to the passengers of the self-driving vehicles. On February 14, 2016, in Mountain View, California, a Lexus SUV sport utility vehicle operating in autonomous mode struck a bus while attempting to drive around a sandbag obstruction in its lane. The vehicle was traveling at approximately 2 miles per hour in a 35 mile per hour zone in the moments leading up to the collision. The car accident was reported to the California Department of Motor Vehicles until May 2018 (Ghorai et al., 2024).

The first fatality involving a self-driving car occurred in the state of Florida, in the United States, on May 7, 2016. In the 2016 incident, an installed advanced driver-assistance program that included brake support was active, but the program required both the driver and the software to agree to initiate braking. Neither performed this task before the car ran into the side of a transport truck. The truck and car traveled perpendicular to each other at the same speed, indicating that the car's sensors detected the truck in a way that the car should have been able to avoid. It is not clear why the car did not brake, and the accident is subject to an ongoing investigation. Courts have struggled with assigning liability in these instances of human-robot interaction in the world. Right now, the general legal status is that the operator of the robot, the one that interacts with the system while it is functioning, maintains the same liability as they would always have, as if the full suite of autonomous systems had not been functioning as designed at the moment of the accident, except for the car manufacturers that have stated that they will be responsible for any accidents that their cars cause while in fully autonomous driving mode. Cheaper insurance prices are predicted to result, as attempts to limit the risk by training and qualifying operators and imposing minimum safety standards. In connection with the development of autonomous or unmanned traffic systems, the expectations for future technological improvements of AIS are that AIS will reduce wildlife-vehicle conflicts, reduce loss of human life, speed up vehicle traffic, and ensure personnel safety (Chougule et al., 2023).

Transparency and accountability are key elements when AI and civil liability in direct as well as in product liability are discussed and affected. Transparency of AI gives insight into the functioning of AI, AI bias, AI functionality, AI policies, AI data, and the systems AI uses. Explainability is a component of transparency and includes information and explanations about the functioning of machine reasoning and machine behavior. It is a characteristic of the interaction between man and the machine. In its purpose to foster public trust and responsible AI development and use, transparency thus also has different dimensions and encompasses societal and individual aspects, as well as economic functions. If something seems biased or unfair in the functioning of machines, it retreats the individual and retards social trust in general. Public trust is one of the core components of a comprehensive AI strategy. It is often voiced that without public trust, AI will not receive the 'social license' to be widely adopted. Trust in AI can be fostered by ensuring the overall accountability of the system's use, by clarifying and explicating for different stakeholders some of the key elements of the AI system, and by taking appropriate safeguards. To this end, enhancing transparency is essential. The current body of laws, policy, and standardization activities are concerned with legal aspects of machines. There is no technical verification available to check the conformance of AI when purchased. The laws on product safety presuppose that machine construction is sound to begin with. The most recent initiative on transparency of AI is concerned with operational aspects. If the laws do not incorporate transparency, one must address liability. It is then that a need for standards arises and the technical verification of the conformity of AI can be prepared (Cheong, 2024).

There is an increasing interest from the legal communities worldwide, as well as from policymakers, in AI and its possible effects on the rules of civil liability. Indeed, several attempts have been made to identify the aforementioned regulatory effects in various jurisdictions. This text takes into account the latest regulatory updates that major policymakers have undertaken to specifically regulate AI after 2021 and aims to illustrate—aside from the main requirements and limitations imposed—from a critical evaluation perspective—to what extent these latest regulatory acts have also determined noteworthy innovations concerning the rules of AI/PPP. The analysis first considers two recent legislative interventions, namely the proposal for a Regulation laying down harmonized rules on Artificial Intelligence, which is subject to an interinstitutional legislative process, and the proposal for a Regulation on machinery products, which has the main purpose of modernizing the regime on placing goods on the market and related procedures (Zaidan; Ibrahim 2024).

Both proposals also contain significant developments relating to the attribution or exclusion of corporate liability for the damages caused by AI system-run machinery, either according to a strict corporate concept or an autonomous conceptualization of AI systems. As far as the US is concerned, another quite remarkable contribution was recently offered by the Biden-Harris Administration, whose Executive Order on the Use of Artificial Intelligence contains not only provisions concerning the guiding principles for AI development and use by federal agencies, but also principles aimed at protecting privacy and civil liberties, as well as establishing a surveillance of AI-related technical and ethical standards that are developed in the private sector nationally and internationally (Dotan, 2024).

Indeed, some articles are of interest in the field of corporate responsibility for the damages caused by AI systems, as they provide for the establishment of an "AI Safety and Governance Interagency Working Group," which is required to be established within 180 days, and the conduction of regular technology safety reviews through the "Senior Agency Officials for Technology Safety and Ethical Use." Not only is AI regulatory innovation continuously ongoing, but legal scholars, grounded in their peculiar expertise, are encouraged to actively participate in the design of these new tools, offering critical evaluations of the new measures adopted and, wherever deemed necessary, suggesting recommendations to adjust them (Biden, 2023).