The speed offset problem: a case for principled cultural sensitivity in automated vehicle design

Abstract Automated vehicles are widely expected to play a key role in achieving the European Union’s road safety target of ‘Vision Zero’, which aims to eliminate all road fatalities in the European Union by 2050. To achieve this goal, early design strategies for automated vehicles prioritised strict compliance with traffic rules, assuming that rule-abiding behaviour would inherently lead to safer outcomes. Yet rule deviations such as speeding may, counter-intuitively, enhance safety in situations where strict compliance cannot reasonably be expected. Such cases highlight a discrepancy between legal compliance and the informal norms that often govern complex real-world traffic. They also expose an ambivalence in compliance. It is often framed as a necessary precondition for safety, yet in certain contexts, strict compliance can itself generate avoidable risks. This design challenge is what I call the ‘speed offset problem’. The problem becomes particularly evident when examined across different cultural contexts, where expectations around acceptable speeding vary significantly. Building on this, I argue that ‘principled cultural sensitivity’ is fundamental for ethically sound design of automated vehicles. As a normative framework, it demands sensitivity to local driving cultures whilst maintaining firm commitment to core ethical principles. The article develops and applies this framework to the speed offset problem, exemplifying how these guardrails may take shape in concrete cases. Three ethical guardrails restrict the speed offset’s implementation. First, a speed offset is morally permissible only when it increases overall safety. Second, a speed offset must not increase overall safety at the expense of particular groups. Third, a speed offset’s legitimacy must be temporary and revisable. Together, these guardrails ensure that speed offsets are not a concession to unsafe norms but a temporary and ethically guided calibration, bridging current traffic realities with long-term safety goals. The article concludes by outlining implications for engineering practice and regulatory frameworks. 1 Introduction Automated vehicles are widely expected to play a key role in achieving the European Union’s road safety target of ‘Vision Zero’, which aims to eliminate all road fatalities in the European Union by 2050. To achieve this goal, early design strategies for automated vehicles prioritised strict compliance with traffic rules, assuming that rule-abiding behaviour would inherently lead to safer outcomes. Whilst this assumption holds true in many scenarios, it does not adequately account for the full complexity of actual traffic. A particularly illustrative example of this challenge is speeding, a widespread and often socially tolerated form of rule deviation. Some assistance systems explicitly allow users to exceed detected speed limits by a predefined margin, effectively embedding discretionary non-compliance into the system’s behaviour. These speed offsets create a normative tension between legal compliance and traffic safety. Rule deviations such as speeding may, counter-intuitively, enhance safety in situations where strict compliance cannot reasonably be expected. When an automated vehicle brakes in response to a frequently violated speed limit sign, it may provoke rear-end collisions from human drivers who would not anticipate such behaviour. Such cases highlight a discrepancy between legal compliance and the informal norms that often govern real-world traffic. They also expose an ambivalence in compliance. It is often framed as a necessary precondition for safety, yet in certain contexts, strict compliance can itself generate avoidable risks. This design challenge is what I call the ‘speed offset problem’. Whilst the speed offset may appear to be a minor configuration option, it demonstrates how seemingly small technical settings can carry far-reaching ethical implications. These implications become particularly evident when examined across different cultural contexts, where expectations around acceptable speeding vary significantly. Building on this, I argue that principled cultural sensitivity is fundamental for ethically sound design of automated vehicles. To this end, the paper is structured as follows. First, it conceptualises the speed offset problem and the normative tensions it reveals. Second, it examines how this problem manifests differently across cultural contexts, focusing on variation in driving norms. Third, it explores how local driving cultures must shape system behaviour, whilst upholding core ethical principles. The paper concludes by outlining implications for engineering practice and regulatory frameworks, advocating for a shift from formal rule adherence to context-aware, culturally informed automated vehicle design. 2 The need for speed In 1997, Sweden introduced its landmark Vision Zero policy, declaring that no loss of life in road traffic is acceptable. The initiative marked a conceptual shift from blaming individual road users for accidents to holding system designers and authorities responsible for preventing harm (Tingvall 2023). Its influence soon extended beyond Sweden. The European Commission (2021) and the United Nations (2020) have since adopted similar goals, aiming to halve road fatalities by 2030 and to eliminate them altogether by 2050. At the heart of Vision Zero lies the Safe System approach, which recognises that human error is inevitable and that the transport system must be designed to prevent such errors from resulting in serious injury or death (OECD/ITF 2016). It views road safety as a systemic outcome shaped by roads, vehicles, road users, post-crash care, and speeds. To conceptualise the speed offset problem, it is necessary to examine how safety is understood within this framework. This paper assumes that there are at least two factors that are relevant for traffic safety: (i) compliance with traffic rules and (ii) compliance with informal norms. From the perspective of traffic rule compliance, safety depends on keeping speed within the limits prescribed by law. Speed is a decisive factor in traffic safety because it affects both the likelihood of crashes and their severity. When researchers describe speed management as a “task of utmost importance” (Belin and Vadeby 2023, p. 972), they ground their claim in this well-established and hardly debatable relation. This is why most of traffic safety research still concentrates on the individual level, seeking primarily to “manage kinetic energy” (Corben 2023, p. 904). Whilst passive safety measures such as seat belts or helmets are essential, the most effective way to limit crash consequences is to reduce the kinetic energy transferred to the human body. Ultimately, this means reducing speed, which is particularly critical for vulnerable road users such as pedestrians and cyclists, who lack the physical protection available to vehicle occupants. These findings have shaped international policy discussions and were reflected in the 2020 Global Ministerial Conference on Road Safety, which called for default 30 km/h limits in urban areas to reduce fatalities and serious injuries amongst vulnerable road users (Fildes et al. 2023). Posted speed limits thus aim to ensure that vehicles do not travel at speeds that would generate more kinetic energy than a human body can survive. However, policy alone cannot ensure compliance. Lie et al. (2023) point out that drivers frequently disregard basic rules, most notably by speeding, but also through behaviours such as impaired driving, inadequate headway, or lapses in attention. They argue that such violations should be restricted by design, given that existing technologies could already prevent many of them. A prominent example of such technology is the Intelligent Speed Assistance (ISA), which uses cameras and map data to recognise speed limits and alerts drivers when they are exceeded. Some versions also include a control function that limits acceleration. And yet, the driver remains in charge. Alerts can be muted and acceleration beyond the limit is still possible by pressing the pedal more firmly or by deactivating the function altogether. This continued protection of a ‘freedom to speed’ reflects that restricting a vehicle’s maximum speed, or preventing drivers from exceeding posted limits, is a highly emotional issue in some cultures. Measures linked to Vision Zero are often criticised as paternalistic—that is, as unjustified interferences with individual freedom, supposedly for the person’s own good. Such resistance, framed as a fight for freedom, remains a major challenge for supporters of Vision Zero. Hansson (2023) offers a historical account of this debate. Since speeding is not merely a self-affecting behaviour but endangers the lives of others, he concludes that restricting it thereby does not even qualify as a paternalistic measure. Although the people’s resistance may be unjustified, it cannot be ignored. Vision Zero, despite its ambitious title, will require compromises to make gradual progress towards its goal. This points to the second safety-relevant factor: compliance with informal norms. It reflects how drivers align their actions with the flow of surrounding traffic, even when that flow deviates from posted speed limits. This pattern is related to what scholars have called the “speed paradox” (Fleiter and Watson 2006). It describes the seemingly paradoxical behaviour of drivers who acknowledge the risks of speeding but nonetheless exceed speed limits on a regular basis. Since speeding is widespread and early generations of automated vehicles will inevitably encounter it in traffic, it is worth questioning the seemingly unassailable assumption that traffic rule compliance invariably improves safety. What if there are circumstances in which breaking the traffic rules enhances safety more than strict adherence? Can traffic rule compliance become a safety risk? To address this possibility, it is necessary to examine the contexts in which speeding occurs. Speeding is not confined to isolated acts of recklessness but often appears systematically in specific traffic environments, with road design and infrastructure playing a major role. Wide urban boulevards, modernised roads that resemble highways but retain their old speed limits, and empty rural stretches all create conditions that invite drivers to exceed restrictions. It is also encouraged by situational factors. Downhill segments can cause unintentional speeding, whilst the likelihood of receiving a fine further shapes driving behaviour (Edquist et al. 2009; Edwards et al. 2014). Traffic rule compliance decreases when speed limits are perceived as unreasonable. In such cases, rule deviations reflect reactions to limits that appear outdated, inconsistent, or otherwise poorly justified (Gargoum et al. 2016). This finding points to an important ethical dimension of speeding. It cannot be explained solely as individual misconduct but also as a response to rules whose justification is not evident. Traditionally, drivers expressed their disregard for specific limits in the most direct way—by pressing the accelerator. Yet for several decades, manual control has no longer been the only means of determining a car’s speed. When a feature called ‘cruise control’ was introduced in the mid-twentieth century, it allowed drivers to maintain a chosen speed automatically, even when that speed exceeded the limit. A more recent development makes such habitual speeding even easier. Speed offsets denote a deliberate deviation from a detected limit by a given amount, e.g. a + 10 offset with a detected speed limit of 100 km/h would lead to a target speed of 110 km/h. A configurable offset, much like cruise control, preserves user discretion, since the driver sets the target speed within predefined bounds. But its very availability can be read as a nudge towards systematic deviation whilst maintaining the appearance of free choice. Even when users configure the offset themselves, questions of liability and accountability already arise, since the system explicitly enables a rule violation that conventional controls merely permit. The concern intensifies when configurability is removed and the system itself applies a fixed offset, leaving drivers with no real alternative. In such cases, engineers exercise discretion on behalf of users. The vehicle may then violate speed limits automatically and without explicit consent. At this point, the practice amounts to a law violation by design. It raises questions about transparency, explainability, and the redistribution of responsibility from driver to manufacturer. Although these ethical and legal implications are substantial, they cannot be addressed in detail here. Accordingly, this paper uses the term speed offset in a general sense to include both configurable and embedded forms. The speed offset function can be seen as an admission that automated vehicles must adapt to the prevailing traffic flow to some extent. Traffic is not strictly based on law and can at times appear chaotic. Whils individuals might increase their own safety by driving significantly slower than the surrounding flow, this uncommon behaviour may provoke risky overtaking attempts or rear-end collisions. At the system level, an act that enhances individual safety can therefore result in a reduction of overall safety. At the same time, traffic safety research often relies on observational data, meaning its findings are primarily correlational and context-dependent, which makes causal claims difficult to establish with confidence. This complicates predictions about the precise effects of specific design choices. The ethical conclusions in this paper should therefore be read as conditional on current evidence and open to revision as our understanding of safety dynamics evolves. Despite these caveats, addressing the normative tension between compliance with traffic rules and compliance with informal norms is crucial to the ethical design of automated vehicles. An exclusive focus on the management of kinetic energy and thereby strict traffic rule compliance risks overlooking these collective outcomes. Whilst both traffic rule compliance and the compliance with informal norms contribute to preventing harm, they can pull in opposite directions. Current automation strategies tend to privilege the former, assuming that traffic rule compliance equates to safety, whilst overlooking how rigid rule-following can itself generate new risks. The complexity of the traffic system must therefore be acknowledged in any serious assessment of safety. The current reality of traffic has led developers of automated vehicles to move away from the ideal of strict rule abidance and towards a concept of a model driver. The aim is not simply to ask what the law prescribes but to model what a human driver would do in a given situation. As Schöneburg and Baumann (2023, p. 738) note, this presupposes “that the algorithm is configured for cooperative behavior—comparable with a considerate driver”. A closer look at this concept reveals another source of variation to which automated vehicles must adapt. The challenge is not only mixed traffic in general but also the cultural settings in which it unfolds, since the norms of driving behaviour vary considerably across cultures. It is therefore crucial to ask from which cultural context the model driver is derived and whether it can or should be applied in contexts that follow very different logics of driving. 3 Driven by culture In the previous section, we saw that traffic safety may depend not only on compliance with traffic rules but also on compliance with informal norms. These informal norms, however, are not universal. The idea of a model driver implies a cultural frame of reference. What counts as model behaviour differs significantly across cultures and is shaped by distinct driving norms and social expectations. To understand how the speed offset problem unfolds globally, we must therefore look beyond technical design and examine the cultural dimension of driving itself. This section explores how these different informal norms create distinct moral and practical contexts for automated vehicles, laying the conceptual groundwork for the discussion that follows. Driving a car has always been more than a technical act. It expresses a set of cultural values. As Kröger (2016) argues, the automobile embodies modernity’s ideal of freedom, yet this freedom carries the risk of harm. This tension led people to associate driving with two contrasting values: the joy of driving powerful cars and the safety of getting from A to B efficiently. Automation aligns largely with the latter, neglecting the symbolic appeal of freedom and risk that once defined driving. Kröger (2016, p. 65) calls this shift not only a technological but above all a cultural one—a “reinvention of the car”. The same tension now reappears in driver assistance systems. Whilst technologies such as ISA discourage drivers from exceeding speed limits, others like the speed offset facilitate it. Their coexistence invites an intercultural question: how do cultures differ in what they expect from a car, and which values do they prioritise? Road traffic constantly involves other drivers, cyclists, and pedestrians who share the same space and bear the consequences of each other’s actions. To capture how people actually behave in traffic, Zaidel (1992) developed the concept of a “culture of driving”. For him, “[e]ach driver is an individual influenced by the social environment consisting of other road users, general social norms, traffic-related rules of conduct, and their representations” (Zaidel 1992, p. 585). Driving as a socially embedded practice is characterised by the fact that each driver’s behaviour both shapes and is shaped by others. It is through imitation, communication, and conformity to perceived norms that a functioning traffic flow emerges. Zaidel’s account can be deepened through Zoller’s (2017) concept of “skilled perception”. For Zoller, skill is not merely a capacity to act but a way of perceiving elements of the world inaccessible to the unskilled. A skilled driver, for instance, perceives traffic as a meaningful field of possibilities rather than a stream of sensory data. Although Zoller does not frame this in cultural terms, his phenomenology of skill invites such an interpretation. What drivers learn to notice or ignore reflects the perceptual habits cultivated by a given driving culture. Färber’s (2016) analysis complements this picture by shifting from perception to interaction. According to him, road traffic is a “self-organizing, chaotic system” (Färber 2016, p. 125) in which drivers continuously negotiate through eye contact, gestures, and anticipatory movement. He notes that “[a]nyone who has visited another country as a tourist, whether as a driver or a pedestrian, has experienced for themselves how drastically informal rules, culturally specific modes of behavior, and communication between road users affect traffic” (Färber 2016, p. 126). Confronted with such diversity, drivers often experience initial irritation before gradually adapting and internalising new norms. Färber illustrates these contrasts vividly. In southern Europe, drivers tend to merge into a lane by accelerating into gaps, trusting that others will yield, whereas in central and northern Europe drivers typically wait for clear feedback, such as eye contact, a nod, or a visible reduction in speed. The same manoeuvre that signals competence in one context can appear reckless or inconsiderate in another, revealing how expectations about interacting in traffic are culturally shaped. This cultural influence seems to extend to speeding behaviour. Empirical studies suggest that compliance with speed limits varies widely across countries. In a comparative analysis of five nations, He et al. (2023) found that most drivers exceeded speed limits in Italy, Japan, and the United States, whereas rates were much lower in China and moderate in Germany. Such cross-national comparisons must be read with caution. Variations in road type, sample composition, and data sources may affect results as much as cultural norms. What appears as a difference in driving culture may therefore at least partly reflect differences in data selection and traffic context. Moreover, driving culture varies not only between countries but also within them. Urban and rural areas, different generations of drivers, and different road types within the same area can exhibit notably different informal norms. This micro-level variation complicates any attempt to define a single national driving culture. An even more granular adaptation may be necessary. Recognising such variation in driving culture raises a philosophical challenge. If cultures differ in what they consider safe or competent driving, can automated vehicles be designed according to universal ethical standards? It is therefore important to stress that acknowledging cultural diversity does not need to entail accepting moral relativism in all of its interpretations. Moral relativism can be understood in several ways. Broadly speaking, it corresponds to “an empirical thesis that there are deep and widespread moral disagreements and a metaethical thesis that the truth or justification of moral judgments is not absolute, but relative to the moral standard of some person or group of persons” (Gowans 2021). Note, however, that the descriptive thesis, i.e. that different cultures have different driving norms, does not already entail the metaethical thesis, i.e. that there is no objective standard by which ethical norms can be evaluated. We should therefore distinguish between surface-level cultural practices and deeper moral principles. Whilst it may well be true that cultures differ in terms of their informal norms regarding speeding, these differences could just reflect differing implementations of what are the same, shared underlying moral values or principles, such as harm prevention and efficient coordination. Therefore, this analysis proceeds on the assumption that cultural variation occurs primarily at the level of informal driving norms rather than at the level of fundamental moral principles. And yet, cultural diversity will remain a crucial challenge for automated vehicle design. The difficulty lies in consistently finding and maintaining the balance between cultural sensitivity and ethical universalism. This challenge extends the normative tension outlined in the preceding section. The earlier contrast between compliance with traffic rules and compliance with informal norms now reappears in cultural form, as a tension between universal ethical principles and diverse driving cultures. The next section introduces a framework, which I call ‘principled cultural sensitivity’, to address this tension, offering a way to approach the speed offset problem on a global scale. 4 Steering between norms Building on the preceding analysis, this section develops the framework of principled cultural sensitivity, examining how automated vehicle design should reflect local driving cultures without compromising core ethical principles. It proceeds as follows. First, it argues that automated vehicle design should reflect the informal norms of the respective driving culture. Second, it demonstrates that this cultural sensitivity should be constrained by ethical guardrails. Third, it applies the framework to the speed offset problem to exemplify how these guardrails may take shape in concrete cases. 4.1 The case for cultural sensitivity To understand why adaptation is normatively required, we must first examine the practical and moral reasons for aligning automated vehicle behaviour with local driving cultures. Winner (2016) already identified individualisation as a future research priority for driver assistance systems. The success of intermediate automation stages depends not only on technical performance but also on how systems align with users’ preferences and cultural expectations. Locally adapted assistance systems would therefore be required “both for market success and also for further increased traffic safety”, since the “location-specific character of the assistance functions will emerge as a new challenge” (Winner 2016, p. 1578). Lin (2016) makes a similar point, adding further pragmatic reasons for such adaptation. According to him, automated vehicles that follow rules too rigidly risk being exploited by human drivers who can predict their overly cautious reactions, “e.g., by cutting in front of it, knowing that the automated car will slow down or swerve to avoid an accident” (Lin 2016, p. 81). He warns that overly conservative driving may even make them “a traffic hazard or trigger road-rage in human drivers” (Lin 2016, p. 80). However, his point ultimately extends beyond pragmatism. He shows that ethical choices are embedded in the design of automated vehicles. The design process therefore improves when these assumptions are made explicit and shaped through dialogue with the broader community, rather than within what Lin (2016, p. 75) calls a “moral bubble”. This insight applies equally to cultural diversity. Designing automated vehicles within a narrow cultural frame risks reproducing local assumptions as if they were universal. Global design processes must therefore recognise that moral expectations and value hierarchies differ across cultures. A system that embodies only one culture’s view of acceptable driving would be unjust towards those whose moral perspectives it excludes. Although such a reductionist approach may appear attractive for its savings in time, cost, and resources, it comes “at the expense of gross inadequacy” (Smilansky 2022, p. 117). No single moral framework can capture the range of perspectives that shape real-world driving behaviour. Yet exclusion is not the only problem. Other driving cultures must be recognised through fair and inclusive procedures. To determine what people truly value, they must be directly involved. Participatory approaches can help avoid speculative or paternalistic assumptions about what a given culture expects an automated vehicle to do—or to refrain from doing. Henrich (2020) shows that cultural practices cannot be treated as mere preferences but are rooted in historically evolved and deeply entrenched moral expectations. He emphasises that much of what Western societies take to be universal moral reasoning is rooted in a historically specific psychological profile. This cultural embeddedness calls for caution against extrapolating from a Western perspective to humanity at large. Dealing with cultural diversity in local driving norms effectively requires considering at least two options for where to locate universal principles. One option would be to acknowledge cultural diversity at the level of informal norms whilst still insisting on a single, universal way of addressing them. This approach has the advantage of clarity, but it flattens meaningful cultural differences by treating different driving practices as if they all warranted the same regulatory or technical treatment. The second option would recognise that not only informal norms vary, but also that there are differing reasonable views about how to address this variation. The universal claim then shifts from a first-order normative ethical level to a meta-ethical level. Rather than prescribing one substantive response to diverse informal norms, we would commit to navigating this tension from a position of moral uncertainty. Bhargava and Kim (2017, p. 6) describe it as a situation in which an agent “has access to all (or most) of the relevant non-moral facts, including but not limited to empirical and legal facts, but still remains uncertain about what morality requires of her”. They argue that programmers need a “meta-normative framework” (Bhargava and Kim 2017, p. 7) for making rational choices under such conditions. Rather than committing to a single moral theory, programmers should weigh alternative principles according to their plausibility and the seriousness of their moral implications. Whilst identifying best practice remains complex, using such frameworks already demonstrates moral concern and helps reduce the “risk of wronging other human beings” (Bhargava and Kim 2017, p. 14). Applied to global automated vehicle design, moral uncertainty extends beyond abstract ethics. It concerns how systems should act amid diverse cultural expectations of proper driving. The question is how far this adaptation to different cultures should go. Taking it to its extreme, Smilansky (2022, p. 123) introduces what he calls “Designer Ethics”, noting that “[p]eople’s views are pro tanto crucial determinants of how, morally, they ought to be treated”. This line of thought implies that a utilitarian driver could, in principle, be considered a more appropriate ‘sacrificial victim’ than a deontological one. Translated to the cultural level, this reasoning raises a difficult question for automated vehicle ethics: If a society places lower value on safety and higher value on risk or freedom, to what extent should automated vehicle design mirror that hierarchy? Put more provocatively, if members of a given culture place little value on their own safety, are we obliged to value it more than they do? In response to this plurality, Smilansky (2022, p. 125, emphasis in original) foresees the emergence of “different AV systems, implementing different preferences and assuming a different ranking of the pertinent values”. He envisions, for example, a Western European model centred on public ownership and collective safety, and a North American model emphasising private ownership, customisation, and differentiated safety levels. Other configurations may lie between these extremes. Smilansky does not view this diversity as a problem but as an opportunity to learn. He sees it as an experiment in applied normative pluralism. Whilst his approach is “completely at odds with standard practice in the universal application of the treatment of persons in such contexts” (Smilansky 2022, p. 127), it illustrates how these unprecedented situations can give rise to correspondingly novel forms of moral reasoning. What sometimes appears as novel moral reasoning may, in fact, reflect perspectives long present in non-Western traditions but often overlooked in global debates on technology. Hongladarom and Novotný (2022, p. 99) illustrate this with their metaphor of the automated vehicle as a novice driver attending a drivers’ school to “learn to share the roads with human drivers in the given cultural and legal context”. For them, such learning includes “how to become a fully functional member” (2022, p. 102) of the respective driving culture. They caution that “a driver who strictly follows the speed limit might obstruct the flow of the traffic and the situation could become rather dangerous” (2022, p. 104). For them, being culturally aware is a part of being morally good. Recognising these differences across driving cultures calls for a flexible approach to automated vehicle design. Such flexibility is compatible with different moral frameworks, although they realise it differently. Within utilitarianism, the contrast between rule- and act-utilitarianism becomes relevant. Bennett (2022) argues for a rule-utilitarian approach, drawing on Goodin (1995), who stresses that rules have a distinct advantage in the realm of public policy because they are easier to communicate, to learn, and to apply. Yet, rule-utilitarianism is not thereby committed to rigid adherence. Hooker (2000) argues that the optimal set of rules should include a disaster prevention clause that permits deviations when strict adherence would produce significantly worse outcomes. Act-utilitarianism, for its part, incorporates flexibility directly through its case-by-case structure, evaluating individual actions by their expected outcomes. Occasional deviations from traffic rules may then be ethically permissible if they lead to greater overall safety. Deontology accommodates flexibility through a structurally similar mechanism. In the broader field of robot ethics, Talbot et al. (2017) examine several approaches, including a form of moderate deontology, which holds that whilst it is generally wrong to violate rights, such violations can be permissible if they produce extremely good consequences or avert extremely bad ones. At the centre of this argument is what the authors call the “dropout threshold” (2017, p. 265, emphasis in original). Once it is reached, the rule’s prohibitive force drops out of the moral calculus. The challenge lies in defining that cut-off point. In the context of automated vehicles, speed limits exist to protect drivers and others in traffic. Whilst individuals may sometimes choose to trade off this safety for other values, they cannot ignore the effects their violations impose on others. Under the form of moderate deontology highlighted above, the likelihood of harm becomes crucial for the moral assessment. When violating a rule carries minimal risk or even contributes to greater overall safety, limited non-compliance may be permissible. The dependency on probabilities and empirical studies in moral programming highlights a parallel difficulty at the procedural level, one that Rawls (1999, §14) characterises as imperfect procedural justice. Whilst the desired outcome is both identifiable and normatively desirable, strict adherence to formal rules does not reliably produce it. Speed limits are designed to promote traffic safety, but their rigid enforcement may at times generate new risks. The speed offset problem exemplifies this indeterminacy. Even when rules are followed, safety is not necessarily secured, because the actual conditions of traffic dynamics introduce elements of unpredictability. Achieving safety under these conditions requires a more flexible approach that acknowledges that traffic rule compliance alone may not secure the intended outcome. All these frameworks accommodate flexibility in different ways. Rule-utilitarianism locates it in an exception clause for cases where strict adherence would be self-defeating. Act-utilitarianism locates it in the basic structure of moral evaluation itself. Moderate deontology locates it in a threshold at which the prohibitive force of a rule gives way to consequentialist considerations. What matters for the present argument is that none of these frameworks rules out the kind of context-sensitive flexibility that culturally adapted automated vehicle design requires. The case for such flexibility therefore does not depend on a commitment to any particular framework. What it does depend on is attention to the empirical realities that shape safety outcomes, including the cultural variation examined above. Given that driving cultures differ significantly and embody distinct moral expectations and value hierarchies, and that automated vehicles designed without regard to these norms may compromise safety and impose the norms of one culture upon others, it follows that automated vehicle design should reflect the informal norms of the respective driving culture. 4.2 Why cultural sensitivity must be principled Yet the very process of cultural adaptation carries its own risks. They call for ethical principles that act as guardrails. The need for principled cultural sensitivity becomes apparent once we distinguish between the different ways in which cultural adaptation can be justified. Such justifications can be grouped into three types: pragmatic, cultural, and normative. The following discussion examines them in turn, outlining the shortcomings of the first two and why a normative approach does not share them. Pragmatic justifications offer the weakest reasons for cultural adaptation. They rest on practical considerations such as improving system performance, user acceptance, or market success. Take, for example, the claim that automated vehicles must adapt to local norms to function within mixed traffic or that users will reject systems diverging too much from prevailing driving styles. Whilst such concerns are legitimate from an engineering or business perspective, they do not confer moral legitimacy. As Berkey (2022) points out in regard to the prioritisation of vehicle occupants’ interests, appeals to consumer preference or feasibility cannot replace ethical reasoning about risk distribution. Blindly following market forces may even worsen inequalities and reinforce the very practices that ought to be questioned. Pragmatic adaptation may thus be necessary for feasibility, but it cannot by itself justify a moral choice. Gerdes and Thornton (2016), for instance, employed a combination of deontological and consequentialist elements to determine the appropriate behaviour of automated vehicles. In their account, traffic rules are not presented as strict constraints. Instead, given their frequent violation, they are treated as one factor in a cost function, balanced against considerations such as mobility needs in the form of smooth traffic flow. And yet, just because “humans tend to treat these laws as guidelines as opposed to hard and fast rules” (Gerdes and Thornton 2016, p. 97), we cannot derive any normative weight from this empirical fact. Doing so would amount to an is-ought fallacy. Capturing current human behaviour on the roads and embedding it in automated vehicles may be a pragmatic way to adapt to mixed traffic and facilitate market penetration, but such an approach does not provide a moral foundation for this behaviour. It remains a pragmatic solution outside of the moral dimension. The fact that humans “expect these sorts of actions from other humans” (Gerdes and Thornton 2016, p. 97) likewise fails to qualify as a moral reason for designing software to intentionally violate traffic laws. Such an approach even risks eroding core societal values. In his paper on the politics of self-driving cars, Pagallo (2022) illustrates this danger with an effect observed in the context of privacy expectations. In the United States, privacy law relies on the idea of a ‘reasonable expectation of privacy’ which combines an individual’s belief that a situation or location is private with society’s recognition of that expectation. This standard assumes that both individuals and society have developed a stable set of privacy expectations. Pagallo (2022, p. 168) points out that this assumption is debatable, since “technology can dramatically change those expectations”. A similar problem can arise when speed offsets reshape expectations about proper driving speed. Becoming accustomed to cars that break the law by design may further erode the social acceptance of speed limits and undermine their credibility. This effect differs from the effect individual rule violations have. When the violation is embedded in the design of vehicles, its official and seemingly legitimate character can function as a false authority on what counts as a proper speed. The core shortcoming of pragmatic justifications is that they treat existing behaviour and expectations as if they carried independent moral weight, erasing the distinction between what people do and what is morally justified. Cultural justifications move beyond mere pragmatism. If, as this paper assumes, cultural practices reflect differing implementations of shared underlying moral values, then such practices do carry prima facie moral authority. Their authority derives from a structural feature of the problem they address. Values such as harm prevention and efficient coordination do not uniquely determine concrete driving practices. The same value can be realised through different conventions. Selection amongst them is required but not dictated by the value itself (cf. Lewis 1969; Marmor 2009). Driving cultures are, in this sense, conventional responses to a problem the underlying values alone leave underdetermined. Their prima facie authority rests on this function, not on their mere facticity. However, prima facie authority is defeasible. An implementation can be a more or less faithful realisation of the values it aims to serve. The critical question, then, is whether a given practice actually advances those values or only appears to do so. Appealing to shared cultural practices or collective preferences may seem morally richer, yet the fact that a culture collectively endorses risky or harmful behaviour does not make it ethically acceptable. Entire communities may normalise speeding, close following, or disregarding vulnerable road users. Hansson (2023) argues that limiting speed is not a paternalistic measure because the decision to speed affects not only the driver but also others who are put at risk against their will. On an individual level, this reasoning is persuasive. On a cultural level, it weakens, since a society may claim the right to decide for itself what level of danger it finds acceptable. With no ‘outsiders’ left to protect, one might be tempted to adopt a live-and-let-live attitude. However, at least one outsider remains: the company that designs and deploys the system. Is it permissible for the company to be neutral towards the moral choices of its customers? Or does its responsibility extend further, obliging it to prevent avoidable harm even when such harm is culturally normalised? Hansson (2023, p. 235) criticises “the absurdity of valuing options for moral deliberation higher than the avoidance of human suffering”. Companies should therefore carefully consider which options their technologies ought to enable and which they have good reason to exclude. The core shortcoming of cultural justifications is that they treat prima facie authority as conclusive, taking the cultural embeddedness of a practice as sufficient warrant without asking whether it faithfully realises the values it claims to serve. Normative justifications do not share these shortcomings. They do not treat existing behaviour and expectations as self-sufficient sources of moral guidance, nor do they treat the cultural embeddedness of a practice as conclusive moral authority. Instead, cultural adaptation can be grounded in ethical principles that retain their validity across contexts. Such principles provide the criteria for evaluating whether specific cultural practices faithfully realise the shared values they claim to serve, distinguishing legitimate forms of cultural sensitivity from arbitrary accommodation. They help identify fundamental moral requirements that remain non-negotiable across cultures and define the ethical minimum that must not be compromised, even when automated vehicles are tailored to local conditions. Given that automated vehicle design should reflect the informal norms of the respective driving culture, and that flexibility without ethical principles risks eroding moral standards, it follows that cultural sensitivity must be constrained by ethical guardrails. 4.3 Ethical guardrails for speed offsets How these ethical guardrails apply to specific cases becomes tangible when considering the speed offset function. The following discussion explores three guardrails that should guide its implementation: overall safety, fairness, and provisionality. Together they show how automated vehicle design can be responsive to local driving cultures whilst remaining constrained by ethical principles. The first guardrail concerns the foundational value that no cultural adaptation in road traffic may compromise: a speed offset is morally permissible only when it increases overall safety. Overall safety is understood here in terms of the aggregate frequency and severity of incidents across all road users within a given traffic environment. An increase in overall safety corresponds to a reduction in these figures. The underlying idea of harm reduction is supported across all prominent ethical theories. Whilst safety outcomes depend on empirical realities, these realities also carry normative weight. Understanding this relationship without collapsing into a simple is-ought fallacy requires an empirically informed ethical approach. Bennett (2022, p. 193) emphasises that “a key aspect of taking a utilitarian approach is that the judgments which utilitarianism produces depend on empirical facts”. To achieve the greater good, one must have sufficient empirical knowledge to assess situations reliably and identify how utility can best be maximised. He argues that utilitarian judgements need a “firm grasp of the facts” (Bennett 2022, p. 193). One fact is that speed limits are frequently violated. The prevalence of such violations is not in itself a reason for claiming their moral rightness, but it is an important fact for an adequate assessment of the traffic situation. It becomes indirectly morally relevant when large deviations from prevailing driving practice introduce new risks, thereby diminishing the very value of safety that speed limits are meant to protect. What is at issue here is not the legitimacy of traffic signs as such but the assumption that compliance with them invariably increases safety. Yet assessing overall safety is not straightforward. Whilst traffic safety research leaves little doubt that higher speeds generally increase risk, there might be systemic conditions under which moderate speed offsets could, paradoxically, enhance overall safety. At the individual level, a speed offset is difficult to defend. If an accident occurs during a system-induced speeding, questions of moral and legal responsibility inevitably arise. From a systemic perspective, however, moderate speed offsets could more accurately reflect the prevailing driving culture and thereby increase public acceptance of driver assistance systems. The safety benefits of these systems cannot unfold fully as long as most drivers rarely or never use them. Because such functions are often offered as part of larger assistance packages, a speed offset feature could make the entire package more attractive, indirectly promoting lane keeping and distance management functions that are generally associated with safety gains. This argument rests on two key premises: (i) speed offsets increase the acceptance of driver assistance systems, and (ii) these systems, in turn, enhance safety. When speed limits were first introduced, one of the main criteria was drivers’ acceptance. As Belin and Vadeby (2023, p. 974) explain, the guiding principle was that the limit should correspond to the speed 85% of the vehicles are not exceeding. Although additional criteria were introduced for sound reasons, the importance of public acceptance should not be overlooked. The consequences of disregarding this aspect become evident in a historical example discussed by Eugensson and Ivarsson (2023, p. 712). In the early 1970s, a U.S. regulation required seat belt interlock systems that prevented the engine from starting unless drivers had fastened their seat belts. Because the devices were often unreliable, many vehicles refused to start even when drivers had already done so. The resulting public backlash led to widespread buying resistance and the eventual repeal of the mandate. This demonstrates how safety innovations that disregard user experience and acceptance can undermine their own purpose and delay genuine progress. It may therefore be necessary to strike a balance between the individual risks of speed offsets and the broader safety gains from more frequent use of driver assistance systems. However, there seem to be no studies examining whether design features such as speed offsets could have these effects on public acceptance (Koteczki and Balassa 2025). The literature remains focussed on general factors such as trust, knowledge, and perceived usefulness. Moreover, acceptance is also likely to depend on how the legal questions surrounding system-induced speeding are resolved. This suggests that the question of whether allowing moderate speed offsets can increase acceptance remains speculative and in need of dedicated research. Even if the first premise holds, it remains to be shown that the resulting use of driver assistance packages enhances safety. Given that “driver-related factors (i.e., error, impairment, fatigue, and distraction) [are] present in almost 90% of crashes” (Dingus et al. 2016, p. 2636), promoting driver assistance systems seems reasonable. This finding supports the case for mitigating specifically human vulnerabilities. It does not, however, establish that automated systems are less prone to error in general, since automation may introduce new failure modes of its own. The safety benefits may also differ depending on the degree of automation. In assisted driving, where the human driver remains in the loop, safety outcomes depend on the quality of human–machine interaction rather than on the system’s autonomous performance alone. The comparative safety effects of these different configurations warrant further empirical investigation. Additionally, although studies consistently associate driver assistance systems with substantial reductions in crash frequency and severity, it remains unclear whether these reductions result from assistance packages whose usage rates would increase if a speed offset feature were introduced. For instance, statistics that include functions such as Autonomous Emergency Braking (AEB) may overstate the safety impact of driver assistance packages, as AEB can operate independently of other features such as speed or lane-keeping assists. The bundled nature of such functions complicates causal analysis, since it is rarely clear which component contributes most to observed improvements. Moreover, drivers who readily adopt advanced assistance packages may already differ in their risk attitudes, which itself could account for part of the measured effect. If the goal is to enhance road safety worldwide, the greatest potential gains lie in regions where fatalities remain highest. According to the World Health Organization (2018), most road fatalities occur outside of Europe, with Africa and South-East Asia disproportionately affected. However, two considerations limit the extent to which driver assistance systems—including speed offsets—can address this disparity. First, automated vehicles and advanced assistance systems may remain financially inaccessible to large parts of the populations most affected by road fatalities. Second, high fatality rates in these regions may be attributable to factors over which driver assistance systems have very limited influence, such as deficient road infrastructure, absent lane markings, lack of crash barriers, or inadequate post-crash care. These structural conditions suggest that the safety benefits of speed offsets and related assistance functions are primarily relevant in traffic environments with well-developed road infrastructure where the prevailing flow of traffic systematically exceeds posted speed limits. This cautions against invoking global fatality data to justify features whose practical benefits may be confined to wealthier markets. Yet the limited global applicability of speed offsets does not diminish the broader obligation to pursue road safety worldwide. As Eugensson and Ivarsson (2023) stress, vehicle manufacturers bear responsibility to ensure that safety technologies are implemented globally. Allowing unequal safety standards would contradict the principle that every human life has equal value, regardless of where one happens to live. A genuine commitment to reducing accidents worldwide must direct attention to where the most lives can be saved. Adapting to the realities of driving cultures with the highest road fatality rates may be necessary to achieve the traffic safety goals adopted by the United Nations. A clearer understanding of these realities requires analysing national accident statistics, such as GIDAS in Germany, CISS in the United States, and CIDAS in China. Examining this data can reveal the main causes of crashes and whether features such as speed offsets and related driver assistance systems could help mitigate them. Still, overall safety is a necessary but not sufficient condition for the moral permissibility of speed offsets. Aggregate analysis must not obscure how safety gains and residual risks are distributed amongst different participants in traffic. Therefore, the second guardrail concerns fairness: a speed offset must not increase overall safety at the expense of particular groups. Why should fairness function as a separate guardrail rather than be incorporated into the assessment of overall safety? The reason is structural. Aggregate safety metrics can mask how gains and residual risks are distributed amongst different road users. A net reduction in crash frequency across a traffic environment can coexist with a significant increase in risk for a particular subgroup, such as pedestrians or cyclists. If fairness is treated merely as one variable within the safety calculus, it risks being outweighed by sufficiently large aggregate gains – which is precisely the outcome the separate guardrail is designed to prevent. This constraint finds support across multiple ethical traditions. Distribution-sensitive forms of utilitarianism object to aggregate improvements achieved at the cost of severe harm to identifiable groups. Deontological approaches hold that individuals must not be instrumentalised for collective benefit. A pedestrian’s exposure to increased risk cannot be justified solely by pointing to reduced risk for vehicle occupants. Rawlsian frameworks direct attention to the position of the worst-off, requiring that design choices do not further disadvantage those already most vulnerable in traffic. Rather than committing to one of these theories, this paper treats fairness as a convergence point: a principle whose force is recognised across the most prominent ethical theories. What may vary across cultures is the specific operationalisation of fairness–whether priority is given to collective welfare, individual rights, or the position of the worst-off. Yet the core commitment that aggregate safety gains must not come at the disproportionate expense of identifiable groups holds across these perspectives. This raises a further question: at what point do differences in risk exposure become morally relevant? Fairness cannot mean perfect equality, since traffic always involves some asymmetry of risk, but it does require a threshold below which disparities remain acceptable. Determining where this threshold lies is both an ethical and an empirical task. Without such calibration, even minor redistributions of danger may appear negligible in data yet remain morally significant for those most exposed. A more pragmatic way to address this challenge is to evaluate fairness by scenario, since the density of vulnerable road users predictably varies across contexts. The applicability of speed offsets is inherently context-dependent. They are primarily relevant in traffic environments where the prevailing flow systematically exceeds posted limits, most commonly on highways and comparable road types. In low-speed environments such as residential streets and urban areas with high pedestrian density, the primary safety strategy should be speed reduction, not flow adaptation. To protect vulnerable road users it might, for instance, be mandatory that speed offsets remain minimal or be entirely prohibited in school zones or in close proximity to a playground. These scenarios can be identified through a combination of map data and traffic sign detection. Moreover, geo-fenced areas could automatically impose strict speed limits at specific times of day (Hansen et al. 2025). Within these constraints, cultural sensitivity is permissible and necessary. The goal is to strike a balance that acknowledges diversity whilst upholding a shared commitment to prevent excessive harm and promote fairness. Striking this balance is an ongoing task. Driving culture and the composition of its participants will continue to change as automation advances. Ethical evaluation must therefore remain sensitive to these changing circumstances. This calls for a third guardrail: a speed offset’s legitimacy must be temporary and revisable. This paper rests on the assumption that automated vehicles will, for the foreseeable future, operate within mixed traffic. In a fully automated traffic system, the need for speeding would naturally disappear. For now, however, principled cultural sensitivity offers a valuable opportunity. Whilst local driving cultures must shape system behaviour, the behaviour of automated vehicles will in turn shape local driving cultures. As Hongladarom and Novotný (2022) suggest, such reciprocal influence could enable gradual cultural transformation. Automated vehicles must initially adapt to local norms to function safely. Yet once they become widespread, their consistent adherence to lawful driving standards could help reform unsafe habits and foster a more safety-oriented driving culture. Zaidel (1992, p. 586) likewise emphasised that drivers are influenced by the collective behaviour of others, describing a potential “snowball effect” through which small, gradual changes could transform the broader environment. Today, this dynamic extends to the interaction between human drivers and automated vehicles. Building on this insight, Edwards et al. (2014) later proposed measures to promote a “traffic safety culture”. Whilst they identified speeding as one of the most persistent challenges that has proven difficult to overcome even through strong enforcement and education, it may also indicate where genuine cultural change could begin. With a gradual, long-term strategy, and through the snowball effect described by Zaidel, automated vehicles could help reshape driving culture step by step towards a safer future. Even if their behaviour still departs modestly from formal limits, a calibrated offset of + 5 km/h remains preferable to human driving cultures that routinely tolerate + 10 km/h. But the same mechanism can work in the opposite direction. If automated vehicles adapt too readily to unsafe norms, they may amplify them. As discussed in Sect. 4.2, this risk is particularly acute because system-level non-compliance, unlike individual rule violations, carries an implicit institutional endorsement that can function as a false authority on what counts as a proper speed. Provisionality can help mitigate this normalisation dynamic. The practical challenge, then, is to ensure that this interactive process of cultural change is guided towards safety rather than away from it. The speed offset should be understood as a time-bound design instrument. Its function is to adapt to the flow of traffic where necessary, yet never to lead or normalise higher speeds. It thereby permits reactive rather than proactive speeding. A speed offset should not imitate every deviation from posted limits. The fact that a single aggressive driver overtakes at excessive speed does not justify treating such behaviour as a cue for adaptation. To prevent such misalignment, safeguards are necessary to ensure that isolated acts of reckless driving do not trigger adaptive responses. In this calibrated sense, the speed offset aligns automated vehicles with the imperfect realities of mixed traffic whilst helping to create the conditions under which such offsets will eventually become unnecessary. The normative force of this guardrail rests on the conditional nature of the offset’s moral justification. A speed offset is permissible only under specific empirical circumstances—most notably mixed traffic in which the prevailing traffic flow significantly exceeds posted limits. These circumstances are not static. As automated vehicle penetration increases and driving culture evolves, the conditions that justified the offset may cease to hold. Maintaining an offset whose justification has expired would mean continuing a practice that is no longer morally permissible. Provisionality is therefore not merely an expression of epistemic humility but an ethical requirement. It ensures that the moral permissibility of an offset is continuously tied to the conditions that ground it. A second ground for provisionality lies in the guardrails themselves. The criteria through which they operate—what counts as ‘overall safety’, what constitutes disproportionate burden on particular groups—are not culturally neutral. Different driving cultures will articulate them differently, and such articulations remain open to contestation. This difficulty cannot be eliminated. Any attempt to constrain cultural adaptation through more abstract principles will reproduce it at another level. Provisionality does not resolve this problem, but it builds the acknowledgement of it into the structure of the framework. It prevents culturally bound articulations of the guardrails from hardening into fixed positions and commits the framework to its own revisability rather than concealing its contingency. The provisionality guardrail also raises practical questions about governance that require explicit treatment. First, provisionality should not designate a single decision-maker. It rather requires a structure in which manufacturers, regulators, and affected communities each play distinct roles. Manufacturers bear the primary responsibility for collecting and sharing data on safety outcomes. Regulators define the criteria under which offsets may be tightened or withdrawn. They retain the authority to set binding constraints. Local communities provide input on the acceptability of changing driving norms through participatory mechanisms, ensuring that neither manufacturers’ commercial interests nor regulators’ standardising tendencies override legitimate local concerns. Second, provisionality demands periodic reassessment as the composition of traffic changes—particularly as the penetration of automated vehicles increases and the rationale for adapting to human driving norms thereby weakens. Third, the burden of justification should lie with those who wish to maintain or extend offsets, not with those who seek to remove them. Tightening or removing offsets could be justified, for instance, when the data indicates that there is a declining speed differential between posted limits and prevailing traffic flow or that there was no measurable safety gain attributable to the offset. Finally, disagreements between manufacturers, regulators, and local communities are to be expected. Such conflicts should be resolved through transparent, evidence-based procedures in which safety data is publicly accessible and independently verifiable. Affected communities should be able to contest both overly permissive and overly restrictive implementations. 5 From principle to practice Principled cultural sensitivity entails two requirements for ethically sound automated vehicle design. It demands sensitivity to local driving cultures whilst maintaining firm commitment to core ethical principles. The first aspect of this framework is adaptive: automated vehicle design should reflect the informal norms of the respective driving culture. The second aspect is constraining: cultural sensitivity must be constrained by ethical guardrails. The discussion of the speed offset function has shown how these two aspects interact in practice. At least three ethical guardrails restrict its implementation. First, a speed offset is morally permissible only when it increases overall safety. Second, a speed offset must not increase overall safety at the expense of particular groups. Third, a speed offset’s legitimacy must be temporary and revisable. Together, they ensure that speed offsets are not a concession to unsafe norms but a temporary and ethically guided calibration, bridging current traffic realities with long-term safety goals. More broadly, principled cultural sensitivity provides a normative framework for automated vehicle design as a whole, shaping how other design values are interpreted and balanced within their cultural context. Having outlined what principled cultural sensitivity requires and how it can be applied to cases such as the speed offset problem, this final section offers a brief outlook on its implementation. Putting this framework into practice entails not only embedding general ethical guardrails into automated vehicle design but also operationalising and testing these concrete guardrails through empirical insight. Therefore, data collection is the central condition for making such moderate adaptations ethically defensible. Only if sufficient data are available to demonstrate that the requirements of overall safety and fairness are met, can a speed offset be considered morally permissible. This requires data on at least four aspects: (i) the relevant driving scenarios, (ii) the characteristics of different driving cultures, (iii) the safety gains achieved through adaptations to traffic flow and through the use of driver assistance systems, and (iv) the redistribution of safety amongst different categories of road users resulting from speed offsets. Once such information is available, it can then be translated into concrete system behaviour. One promising approach to obtain the necessary data is the so-called ‘shadow mode’, in which the automated driving function runs in a virtual, non-intervening mode alongside a human driver, logging sensor data and its own hypothetical actions for later analysis (Junietz et al. 2019). By observing how human drivers negotiate local traffic and comparing it to the system’s planned trajectories, developers can iteratively adapt automated vehicle behaviour to different driving cultures without exposing road users to additional risk during the learning phase. Such learning inevitably relies on extensive sensor data and visual recordings, raising questions about consent, data governance, and the limits of legitimate observation (Kargl et al. 2022). Beyond these data collection challenges, several further issues will determine how principled cultural sensitivity can be implemented in practice. At least two of these have already been touched upon in this paper. Both arise from the characteristics of the different forms of speed offsets introduced in Sect. 2. That distinction was set aside earlier for the sake of conceptual clarity. The first issue concerns the ethical dimension. It asks how configured and embedded offsets must be treated differently regarding transparency, explainability, and the redistribution of responsibility from driver to manufacturer. Even with a configured offset, questions arise about the default setting and how prominently this option can appear in the interface without amounting to a nudge towards unlawful behaviour. Embedded versions raise additional concerns about the design of the human–machine interface. It must be decided how to inform users about potential speed limit violations before they begin their trip, and how such violations are communicated when they occur, to ensure that users can intervene in time to prevent a potentially harmful outcome. The second issue is closely related to these ethical questions and involves the legal implications of speed offsets. Configured versions raise questions about the permissibility of offering users a choice that entails the intentional deviation from traffic rules. When such deviations are embedded in the system’s behaviour, questions arise about the legality of what could be described as a rule violation by design. These cases challenge existing liability frameworks, which typically presuppose human decision-making and individual fault. They require clarification of who bears responsibility when a vehicle’s programmed behaviour, rather than the driver’s discretion, results in a violation of traffic law. In both cases, different implementations reflect different legal and moral risk attitudes. Currently, traffic law in most jurisdictions does not recognise maintaining the flow of traffic as a legal defence for exceeding posted speed limits, which underscores the need for regulatory clarification. The question is whether the law should tolerate certain deviations for the sake of overall safety, or prohibit them categorically to preserve legal certainty. Clarifying how this tension should be resolved will be essential for regulators seeking to align automated vehicle design with the diversity of driving cultures outlined above. The speed offset problem is more than a technical question of how automated vehicles should regulate their speed. It exemplifies the broader normative challenge of embedding ethical reasoning into automated systems that must operate across diverse cultural contexts. Every design choice entails a continuous negotiation between ethical principles and local practices, between rule-based constraint and adaptive judgement. The speed offset problem thus serves as a lens through which the deeper moral and cultural tensions of automated vehicle design become visible. Data availability No datasets were generated or analysed during the current study. References Belin M-Å, Vadeby A (2023) Speed and technology: different modus of operandi. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 971–994 https://doi.org/10.1007/978-3-030-76505-7_37 Bennett S (2022) Algorithms of life and death: a utilitarian approach to the ethics of self-driving cars. In: Jenkins R, Cerný D, Hríbek T (eds) Autonomous vehicle ethics. Oxford University Press, New York, pp 191–209 https://doi.org/10.1093/oso/9780197639191.003.0012 Berkey B (2022) Autonomous vehicles, business ethics, and risk distribution in hybrid traffic. In: Jenkins R, Cerný D, Hríbek T (eds) Autonomous vehicle ethics. Oxford University Press, New York, pp 210–228 https://doi.org/10.1093/oso/9780197639191.003.0013 Bhargava VR, Kim TW (2017) Autonomous vehicles and moral uncertainty. In: Lin P, Jenkins R, Abney K (eds) Robot ethics 2.0: from autonomous cars to artificial intelligence. Oxford University Press, New York, pp 5–19 https://doi.org/10.1093/oso/9780190652951.003.0001 Corben B (2023) Urban road design and keeping down speed. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 903–945 https://doi.org/10.1007/978-3-030-76505-7_35 Dingus TA, Guo F, Lee S et al (2016) Driver crash risk factors and prevalence evaluation using naturalistic driving data. Proc Natl Acad Sci U S A 113:2636–2641. https://doi.org/10.1073/pnas.1513271113 Edquist J, Rudin-Brown C, Lenné MG (2009) Road design factors and their interactions with speed and speed limits. Monash University Accident Research Centre, Victoria, Australia Edwards J, Freeman J, Soole D, Watson B (2014) A framework for conceptualising traffic safety culture. Transp Res F Traffic Psychol Behav 26:293–302. https://doi.org/10.1016/j.trf.2014.03.002 Eugensson A, Ivarsson J (2023) What the car industry can do: Volvo Cars. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 687–725 https://doi.org/10.1007/978-3-030-76505-7_28 European Commission (2021) Commission Delegated Regulation (EU) 2021/1958 of 23 June 2021 supplementing Regulation (EU) 2019/2144 of the European Parliament and of the Council by laying down detailed rules concerning the specific test procedures and technical requirements for the type-approval of motor vehicles with regard to their intelligent speed assistance systems and for the type-approval of those systems as separate technical units and amending Annex II to that Regulation (Text with EEA relevance). Publications Office of the European Union. https://eur-lex.europa.eu/eli/reg_del/2021/1958/oj. Accessed 15 Dec 2025 Färber B (2016) Communication and communication problems between autonomous vehicles and human drivers. In: Maurer M, Gerdes JC, Lenz B, Winner H (eds) Autonomous driving: technical, legal and social aspects. Springer, Berlin, Heidelberg, pp 125–144. https://doi.org/10.1007/978-3-662-48847-8_7 Fildes BN, Lawrence B, Thompson L, Oxley J (2023) Speed-limits in local streets: lessons from a 30 km/h trial in Victoria, Australia. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 881–901 https://doi.org/10.1007/978-3-030-76505-7_34 Fleiter JJ, Watson BC (2006) The speed paradox: the misalignment between driver attitudes and speeding behaviour. J Road Saf 17:23–30 Gargoum SA, El-Basyouny K, Kim A (2016) Towards setting credible speed limits: identifying factors that affect driver compliance on urban roads. Accid Anal Prev 95:138–148. https://doi.org/10.1016/j.aap.2016.07.001 Gerdes JC, Thornton SM (2016) Implementable ethics for autonomous vehicles. In: Maurer M, Gerdes JC, Lenz B, Winner H (eds) Autonomous driving: technical, legal and social aspects. Springer, Berlin, Heidelberg, pp 87–102 https://doi.org/10.1007/978-3-662-48847-8_5 Goodin RE (1995) Utilitarianism as a public philosophy. Cambridge University Press, Cambridge Gowans C (2021) Moral relativism. The Stanford Encyclopedia of Philosophy, https://plato.stanford.edu/entries/moral-relativism/. Accessed 15 Dec 2025 Hansen L, Graupner S-T, Andersson K et al (2025) Geofencing to accelerate digital transitions in cities: Experiences and findings from the GeoSence Project. In: McNally C, Carroll P, Martinez-Pastor B et al (eds) Transport Transitions: Advancing Sustainable and Inclusive Mobility. Springer Nature Switzerland, Cham, pp 89–95 https://doi.org/10.1007/978-3-032-06763-0_13 Hansson SO (2023) Liberty, paternalism, and road safety. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 205–242 https://doi.org/10.1007/978-3-030-23176-7_6-2 He Z, Nanni M, Pappalardo L, Santi P, Ratti C (2023) Speed limit: obey, or not obey? arXiv. https://doi.org/10.48550/arXiv.2307.12189. Accessed 15 Dec 2025 Henrich J (2020) The WEIRDest people in the world: how the West became psychologically peculiar and particularly prosperous. Farrar, Straus and Giroux, New York Hongladarom S, Novotný DD (2022) Autonomous vehicles in drivers’ school: a non-western perspective. In: Jenkins R, Cerný D, Hríbek T (eds) Autonomous vehicle ethics. Oxford University Press, New York, pp 99–113 https://doi.org/10.1093/oso/9780197639191.003.0006 Hooker B (2000) Ideal code, real world: a rule-consequentialist theory of morality. Clarendon Press, Oxford Junietz P, Wachenfeld W, Schönemann V, Domhardt K, Tribelhorn W, Winner H (2019) Gaining knowledge on automated driving’s safety—the risk-free VAAFO tool. In: Waschl H, Kolmanovsky I, Willems F (eds) Control strategies for advanced driver assistance systems and autonomous driving functions. Springer International Publishing, Cham, pp 47–65 https://doi.org/10.1007/978-3-319-91569-2_3 Kargl F, Krontiris I, Weimerskirch A, Williams I, Trkulja N (2022) Privacy protection of automated and self-driving vehicles (Dagstuhl Seminar 22042). Dagstuhl Rep 12:83–100. https://doi.org/10.4230/DagRep.12.1.83 Koteczki R, Balassa BE (2025) Systematic literature review of user acceptance factors of advanced driver assistance systems across different social groups. Transp Res Interdiscip Perspect 31:101486. https://doi.org/10.1016/j.trip.2025.101486 Kröger F (2016) Automated driving in its social, historical and cultural contexts. In: Maurer M, Gerdes JC, Lenz B, Winner H (eds) Autonomous driving: technical, legal and social aspects, Springer, Berlin, Heidelberg, pp 41–68 https://doi.org/10.1007/978-3-662-48847-8_3 Lewis D (1969) Convention: a philosophical study. Harvard University Press, Cambridge, MA Lie A, Tingvall C, Håkansson M, Boström O (2023) Automated vehicles: how do they relate to Vision Zero. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 1057–1071 https://doi.org/10.1007/978-3-030-76505-7_39 Lin P (2016) Why ethics matters for autonomous cars. In: Maurer M, Gerdes JC, Lenz B, Winner H (eds) Autonomous driving: technical, legal and social aspects, Springer, Berlin, Heidelberg, pp 69–85 https://doi.org/10.1007/978-3-662-45854-9_4 Marmor A (2009) Social conventions: from language to law. Princeton University Press, Princeton OECD/ITF (2016) Zero road deaths and serious injuries: leading a paradigm shift to a safe system. OECD Publishing, Paris https://doi.org/10.1787/9789282108055-en Pagallo U (2022) The politics of self-driving cars: soft ethics, hard law, big business, social norms. In: Jenkins R, Cerný D, Hríbek T (eds) Autonomous vehicle ethics. Oxford University Press, New York, pp 159–175 https://doi.org/10.1093/oso/9780197639191.003.0010 Rawls J (1999) A theory of justice: revised edition. Harvard University Press, Cambridge, MA Schöneburg R, Baumann K-H (2023) What the car industry can do: Mercedes-Benz’ view. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 727–754 https://doi.org/10.1007/978-3-030-23176-7_29-1 Smilansky S (2022) Autonomous vehicles and normative pluralism. In: Jenkins R, Cerný D, Hríbek T (eds) Autonomous vehicle ethics. Oxford University Press, New York, pp 114–129 https://doi.org/10.1093/oso/9780197639191.003.0007 Talbot B, Jenkins R, Purves D (2017) When robots should do the wrong thing. In: Lin P, Jenkins R, Abney K (eds) Robot ethics 2.0: from autonomous cars to artificial intelligence. Oxford University Press, New York, pp 258–273 https://doi.org/10.1093/oso/9780190652951.003.0017 Tingvall C (2023) Vision zero: how it all started. In: Edvardsson Björnberg K, Hansson SO, Belin M-Å, Tingvall C (eds) The Vision Zero handbook: theory, technology and management for a zero casualty policy. Springer International Publishing, Cham, pp 245–266 https://doi.org/10.1007/978-3-030-76505-7_8 United Nations (2020) Improving global road safety: second decade of action for road safety 2021–2030 (A/RES/74/299). United Nations, New York Winner H (2016) ADAS, quo vadis? In: Winner H, Hakuli S, Lotz F, Singer C (eds) Handbook of driver assistance systems. Springer International Publishing, Cham, pp 1557–1584 https://doi.org/10.1007/978-3-319-12352-3_62 World Health Organization (2018) Global status report on road safety 2018. World Health Organization, Geneva Zaidel DM (1992) A modeling perspective on the culture of driving. Accid Anal Prev 24:585–597. https://doi.org/10.1016/0001-4575(92)90011-7 Zoller D (2017) Skilled perception, authenticity, and the case against automation. In: Lin P, Jenkins R, Abney K (eds) Robot ethics 2.0: from autonomous cars to artificial intelligence. Oxford University Press, New York, pp 80–92 https://doi.org/10.1093/oso/9780190652951.003.0006 Acknowledgement I would like to thank my colleagues at the Institute for Ethics in Technology at Hamburg University of Technology (TUHH) for their insightful discussions and support. Funding Open Access funding enabled and organized by Projekt DEAL. Author information Authors and Affiliations Contributions JH is the sole author. The author conceived, wrote, and revised the manuscript, approved the final version, and is accountable for all aspects of the work. Corresponding author Ethics declarations Conflict of interest The author is employed by Mercedes-Benz Group AG in the field of automated driving ethics. This article was written in the author’s capacity as a PhD candidate at the Hamburg University of Technology (TUHH). The views expressed are solely those of the author and do not necessarily reflect the views of Mercedes-Benz Group AG. Additional information Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Rights and permissions Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. About this article Cite this article Hölzer, J. The speed offset problem: a case for principled cultural sensitivity in automated vehicle design. AI & Soc (2026). https://doi.org/10.1007/s00146-026-03222-3 Received: Accepted: Published: Version of record: DOI: https://doi.org/10.1007/s00146-026-03222-3