_edited.png)
Attack drones, autonomy and lethal decision-making: the Portuguese case and the ethical limits of technological warfare.
- May 11
- 8 min read
In March 2026, it was reported that the Portuguese Army tested cursor munitions, also known as loitering munitions or "attack drones," during the Strong Impact 2026 exercise, held at the Santa Margarida Military Camp in Constância, with the participation of forces from NATO countries. According to the information released, these systems were developed within the scope of the "Robotics and Autonomous Systems" project, foreseen in the Military Programming Law, in collaboration with the Portuguese company UAVision. The operational logic presented is to search, observe, and only then attack, allowing the operator to monitor the situation in real time, choose the moment to employ force and, in principle, abort or redirect the attack (Cardoso, 2026).
This case should not be treated as an anomaly or a mere technological curiosity. It is a Portuguese example of a broader transformation of contemporary warfare: the convergence of robotics, sensors, artificial intelligence, unmanned systems, smart munitions, and increasingly accelerated decision-making processes. The problem lies not only in the existence of a new weapon, but also in the gradual alteration of the relationship between the soldier, the target, the political command, the defense industry, and the decision to use lethal force.
What are the dangers of autonomous weapons? | The Laws of War | ICRC
International Committee of the Red Cross
It is important to begin with an essential distinction. A cursor munition is not necessarily an autonomous weapon in the strict sense of the term. Many of these weapons still depend on a human operator to confirm the target and authorize the attack. However, they occupy a particularly sensitive intermediate technological zone: they are simultaneously a surveillance instrument, a reconnaissance platform, and a means of destruction. Unlike a classic missile, which is launched against a predetermined target, cursor munition can remain in flight, gather information, track movements, and attack when an operational opportunity arises. It is this fusion of observation and attack that makes its ethical and political significance particularly demanding.
The military promise is clear. Systems of this type can increase accuracy, reduce military exposure, allow for more informed decisions, and prevent more indiscriminate forms of violence. If a cursor munition allows a legitimate military objective to be achieved without resorting to area artillery, wider bombardment, or direct troop approach, there may be a real humanitarian advantage. This possibility should not be ignored. The ethics of war do not demand the automatic rejection of technology; rather, they demand an assessment of whether the technology effectively reduces suffering, preserves the distinction between combatants and civilians, and maintains human responsibility for the use of force.
The risk is that the same technology that promises precision may also lower the threshold for resorting to force. When attacking becomes cheaper, faster, more distant, and seemingly more controlled, the decision to attack may become politically easier and operationally more frequent. Technical precision can thus produce a paradoxical effect: reducing damage in each individual attack, but increasing the overall willingness to resort to lethal force. The central ethical question is not simply whether one weapon is more precise than another. It is whether it alters the decision-making culture that precedes the use of force.
Peter Asaro has argued that lethal decisions require substantive human judgment, precisely because they involve moral responsibility, contextual interpretation, and recognition of the person targeted as a human subject and not merely as an operational object (Asaro, 2013). Amanda Sharkey adds a crucial dimension: human dignity is affected when death is administered by systems that cannot comprehend the moral value of the life they eliminate (Sharkey, 2019). These arguments do not imply that all remote military technology is illegitimate. But they render insufficient the response that “there is always a human operator.” The question is what kind of human presence exists, with what information, at what time, under what pressure, and with what real possibility of refusal.
This is where the distinction between human-in-the-loop and meaningful human control becomes crucial. A human being can be formally in the loop and yet only perform a hasty validation of a technically pre-formatted decision. If the system detects, tracks, suggests, and frames the target, and the operator has only a few seconds to confirm the action, human responsibility becomes more fragile. Santoni de Sio and Van den Hoven argue that meaningful human control requires systems designed to allow moral traceability, adequate understanding, and responsible attribution of the decision (Santoni de Sio and Van den Hoven, 2018). The problem is not just who presses the button. It's whether that gesture corresponds to a full human decision or a formality within an automated chain.
Autonomy in this domain rarely emerges abruptly. It typically progresses in layers: automatic stabilization, assisted navigation, target tracking, pattern recognition, attack recommendation, algorithmic prioritization, and eventually, autonomous target selection. The International Committee of the Red Cross defines autonomous weapons systems as those that, once activated, can select and attack targets without further human intervention. The ICRC has warned that these systems pose serious legal, humanitarian, and ethical risks, particularly when the user cannot predict with sufficient accuracy who will be hit, where, and under what circumstances (ICRC, 2026).
This gradual trajectory is particularly important for the Portuguese case. The publicly available information does not allow us to conclude that the systems tested by the Portuguese Army are lethal autonomous weapons. However, it does allow us to recognize that Portugal is entering a technological ecosystem in which the difference between a remotely operated system, a semi-autonomous system, and an autonomous system may become increasingly less evident to the public. Ethical discussion should occur before standardization, not just after operational integration.
Responsibility also becomes more difficult to pinpoint. In a conventional attack, there are already multiple levels of responsibility: soldier, commander, chain of command, political authority, and the State. In a system mediated by software and sensors, new actors are added: programmers, manufacturers, data providers, system integrators, testing teams, certification bodies, and algorithmic models. Rebecca Crootof has shown that autonomous weapons can cause serious damage without it being easy to identify individual criminal intent, arguing for the need to consider forms of accountability appropriate to automated warfare (Crootof, 2016). This reflection is essential: the more distributed the decision-making, the greater the need for auditing, traceability, and institutional responsibility.
Nationally, this case reveals a Portuguese orientation towards military modernization, NATO integration, and the enhancement of the defense technology industry. The collaboration with UAVision demonstrates that Portugal does not intend to be merely a consumer of external technologies, but also a participant in the development of its own capabilities. This has strategic and industrial advantages, but requires a democratic counterpart: emerging military capabilities, especially when they may involve lethal force, must be accompanied by public mechanisms of proportional scrutiny. This is not about disclosing sensitive operational information, but about defining principles, limits, and responsibilities.
The European Union has sought to assert a normative line based on preserving human control. In 2018, Federica Mogherini stated in the European Parliament that decisions regarding the use of lethal force should remain under human control and that human beings should continue to be responsible for life-and-death decisions (EEAS, 2018). Within the framework of the European Defence Fund, the European Commission has indicated that actions aimed at developing autonomous lethal weapons without the possibility of significant human control over the selection and engagement of human targets are not eligible for European funding (European Parliament, 2024). This position is relevant, but it does not eliminate the internal tension within the EU itself: while asserting ethical limits, the Union promotes strategic autonomy, innovation in defence, robotics, AI, and autonomous systems through its industrial and security instruments.
NATO faces a similar tension, albeit with a grammar more oriented towards responsible adoption. NATO's 2021 Artificial Intelligence Strategy established principles such as legality, accountability, explainability, reliability, governance, and bias mitigation (NATO, 2021). The revised 2024 version reinforced the need to accelerate the adoption of AI in defense, respond to generative AI, improve interoperability, and develop assessment and governance tools (NATO, 2024). The Alliance thus seeks to maintain technological advantage without abandoning a language of democratic accountability. The challenge is whether this language will translate into real operational limits or merely into general principles compatible with a continuous acceleration of military automation.
Political reflection must avoid two extremes. The first is uncritical technological enthusiasm, which sees each new capability as inevitable and desirable. The second is abstract rejection, which ignores the fact that more precise technologies can, in some contexts, reduce suffering and protect lives. Between these extremes lies a more demanding position: accepting that military technology evolves, but refusing to accept that this evolution dispenses with human control, democratic scrutiny, and legal accountability.
Several minimum measures seem necessary. First, meaningful human control over any lethal decision must be preserved, understood not as the mere formal presence of an operator, but as the effective capacity to understand, deliberate, refuse, and answer for the decision. Second, systems that select and attack people without meaningful human control should be prohibited or excluded. Third, the acquisition and testing of emerging military technologies should include independent legal and ethical review, with the participation of experts in international humanitarian law, ethics of war, engineering, AI, security, and fundamental rights. Fourth, technical auditability must exist: whenever a system is used, it must be possible to reconstruct the decision chain. Finally, the State must assume clear institutional responsibility for the systems it acquires, integrates, and employs, without shifting the blame to private suppliers or to technical complexity.
The Portuguese case of cursor munitions does not offer definitive answers. Rather, it offers a starting point. Portugal faces the same question that the EU, NATO, and the United Nations are currently grappling with: how to integrate emerging military technologies without allowing the decision to kill to be progressively transformed into an administrative consequence of sensors, algorithms, and procedures?
A democracy should not measure its technological maturity solely by its ability to develop or acquire advanced systems. It should also measure it by its capacity to recognize that certain decisions require qualified human involvement, political accountability, and clear normative boundaries. Precision does not replace judgment. Distance does not diminish moral obligation. Automation cannot obscure responsibility.
The question remains open: to what extent can a democratic society accept increasingly autonomous weapons without first defining which decisions should always belong to human beings?
Questions for further research
What concrete criteria allow us to distinguish meaningful human control from mere operational validation?
Can a weapon be ethically preferable for reducing collateral damage if, at the same time, it makes the use of force more frequent?
What parliamentary or independent scrutiny mechanisms should accompany the acquisition of military systems with autonomous or semi-autonomous functions?
How should responsibility be assigned when a lethal decision results from the interaction between operator, commander, software, manufacturer, and operational doctrine?
Will the European Union be able to reconcile strategic defense autonomy with a robust ban on autonomous lethal weapons without significant human control?
Is NATO building a substantive model of ethical governance or merely a language of accountability to accelerate the adoption of military AI?
Should Portugal create a national commission for ethical and legal evaluation of emerging military technologies?
References
Asaro, P. (2013) ‘On banning autonomous weapon systems: human rights, automation, and the dehumanization of lethal decision-making’, International Review of the Red Cross .
Cardoso, IS (2026) 'Portuguese army tests “attack drones” in exercise with NATO countries', Euronews , 25 March.
Crootof, R. (2016) 'War Torts: Accountability for Autonomous Weapons', University of Pennsylvania Law Review , 164, pp. 1347–1402.
European External Action Service (2018) 'Autonomous weapons must remain under human control, Mogherini says at European Parliament'.
European Parliament (2024) 'Use of lethal autonomous weapons systems', Parliamentary question E-002645/2024.
International Committee of the Red Cross (2026) 'Autonomous Weapon Systems and International Humanitarian Law: Selected Issues'.
NATO (2021) 'Summary of the NATO Artificial Intelligence Strategy'.
NATO (2024) 'Summary of NATO's revised Artificial Intelligence Strategy'.
Santoni de Sio, F. and Van den Hoven, J. (2018) 'Meaningful Human Control over Autonomous Systems: A Philosophical Account', Frontiers in Robotics and AI , 5.
Sharkey, A. (2019) 'Autonomous weapons systems, killer robots and human dignity', Ethics and Information Technology , 21, pp. 75–87.
Comments