On 25 August 2026, the world’s most prominent humanitarian and diplomatic voices converged on a single, stark message: the era of machines deciding who lives and who dies is no longer a distant scenario, but an imminent risk that demands immediate legal action. In a joint appeal issued in Geneva, UN Secretary?General António Guterres and International Committee of the Red Cross (ICRC) President Mirjana Spoljaric declared,
“We are now dangerously close to crossing a moral red line: the autonomous targeting of humans by machines.”
Their call was not merely rhetorical; it came with a precise policy prescription—
“Negotiations must begin now to urgently adopt a legally binding instrument to regulate autonomous weapon systems with clear prohibitions and restrictions.”
The timing is deliberate. With the seventh Review Conference of the Convention on Certain Conventional Weapons (CCW) scheduled for November 2026, the UN and ICRC are pressing states to move beyond years of technical discussions and open formal treaty negotiations. As Guterres and Spoljaric put it,
“States must show political courage and move beyond incremental discussions towards decisive action.”
The subtext is clear: further delay “will cost lives.”
What autonomous weapons are—and why they matter now
The heart of the issue lies in the definition of
“systems that, once activated, can select and engage targets without further human intervention,”
according to the ICRC. This is not the science-fiction notion of “killer robots” seen in films; rather, it is about real-world technologies ranging from highly restricted defensive systems to more flexible systems able to pursue target acquisition through time and space. According to the ICRC, there are actually three distinct classes of systems that are frequently confused in public debate. First, AI-assisted targeting systems that suggest targets but do not make the decisions themselves. Second, automated defensive systems like missile interceptors on ships working in clearly defined environments against clearly military targets. Finally, autonomous weapons systems (AWS), which, once triggered, are capable of selecting targets and applying force without additional human intervention. It is the latter system aimed at people that needs to be prohibited or heavily restricted, according to the United Nations and the ICRC.
It is due to the reality that the line dividing these two categories is becoming increasingly blurred on real battlegrounds. From Ukraine to Gaza, Sudan to the Gulf, semi-autonomous and AI-powered weapons are used, such as drones which can proceed to strike even without being in communication with the controllers in case of jamming or disruption of connections. Even though there are no reports yet of fully autonomous weapons autonomously engaging targets, the few uses of autonomous weapons for defensive purposes, together with the swift pace of AI integration, raise concerns about the possibility of crossing this threshold very soon.
A two?tier regulatory blueprint
Rather than demanding a complete ban on military artificial intelligence (AI), the joint appeal seeks to establish a two-level regulatory regime that has taken shape through decades of research and diplomatic negotiations led by the ICRC. First, there is a prohibition on the use of AWS for targeting human beings – anti-personnel autonomous weapons. It is argued by the ICRC and UN-oriented non-governmental actors that these weapons should be banned on ethical grounds since they are inconsistent with basic tenets of IHL and respect for human dignity. Second, the prohibition should apply to those autonomous weapons whose effect cannot be reasonably anticipated, predicted, or explained. In today’s rapidly developing field of AI technology, there are a lot of complex autonomous systems whose operation is based on machine-learning models, large amounts of data, and algorithms that even developers are not able to predict and control.
As was pointed out by the ICRC, unpredictable and uncontrollable autonomous systems could violate IHL as soon as the commander would not be able to assess whether the action can be qualified as proportionate, discriminating, and precautionary. Second, there should be strict regulation of all other kinds of AWS. Such a regulation is aimed at setting certain constraints on how, where, and against what kind of objects those weapons may be employed. Specifically, the following elements should be included: limitations regarding targets (only military objectives by nature and not people), duration of autonomous operation, geographical area, and operation context. In addition, there should be human supervision of autonomous operation, thorough testing and evaluation, legal assessment according to Article 36 of Additional Protocol I, and a continuous chain of command.
This blueprint is not abstract. It reflects the ICRC’s October 2025 position paper and subsequent 2026 advocacy, which argue that while existing IHL applies to AWS, it is
“necessary but insufficiently clear for the AI age.”
Hence the push for a new CCW protocol or treaty that codifies bright?line prohibitions and operational guardrails.
Battlefield realities: from theory to practice
The diplomatic push is anchored in observable trends. In Ukraine, both sides have integrated AI?assisted targeting and autonomous features into drone operations, particularly to overcome electronic warfare that severs operator links. In Gaza and Sudan, loitering munitions and armed drones with varying degrees of autonomy have been deployed in dense urban environments where civilian harm risks are high. In the Gulf, the legacy of the 2019 Abqaiq–Khurais attacks and subsequent drone and missile exchanges has accelerated investment in autonomous air and missile defence.
Often referenced as an example of such an event is the incident in Libya in 2020, in which a panel of experts, reported by the UN, identified the use of a loitering munition attacking retreating fighters without specific permission from a human to carry out this mission. While there are some discrepancies regarding the facts of this event, it became a reference point for campaigners who believe that the first event of lethal autonomous systems used against human beings has already happened. The difference between the present-day situation and previous ones is not only the existence of autonomous features but the fact that the element of human control is disappearing. Even if a person is authorized to approve a particular mission, he can receive dozens of recommendations about how to do it within a short period of time. This is what ICRC refers to as the problem of “cognitive overload.”
The diplomatic arithmetic: who wants a treaty, and who resists?
The CCW process has debated LARWs since 2014, but progress has been glacial, constrained by the consensus-based nature of the forum, allowing a handful of militarily powerful states to obstruct any binding measure. Nonetheless, the political winds have changed direction. UN General Assembly resolutions on AWS have been adopted by large majorities: 166 to 3 in December 2024 and 164 to 6 in December 2025, in response to rising global alarm about the threat of a developing arms race, lowered thresholds for use of force, and dangers of miscalculations and rapid escalation.
As of mid-2026, more than 70 states have supported the CCW GGE “rolling text” for negotiations compared to just 42 states in September 2025. In excess of 120 states have expressed support for a new legally binding agreement on LARWs, with regional organizations (CARICOM, the Belén Communiqué from Latin America, and the Freetown Communiqué of ECOWAS) issuing declarations demanding such binding norms. Opponents of the proposed new treaty remain limited to a handful of states with advanced militaries. The United States and Russia have opposed a new legally binding measure, claiming the adequacy of existing IHL provisions and weapons review processes of individual states, and lack of clarity of terms.
The UN and ICRC’s joint appeal is aimed squarely at this impasse. By framing the issue as a moral red line and coupling it with a concrete regulatory blueprint, they seek to raise the political cost of obstruction ahead of the November Review Conference. The message is blunt: if states fail to “open formal negotiations” now, they will be complicit in normalizing a future where algorithms, not humans, choose who dies.
Humanitarian law, accountability, and the moral line
The legal argument rests on core IHL principles: distinction, proportionality, and precautions in attack. Autonomous systems that target humans or operate unpredictably risk violating these principles by design. Machine?learning models trained on biased or incomplete data may misclassify civilians as combatants; adaptive algorithms may behave differently in novel environments; and complex sensor fusion may fail under jamming or spoofing. The ICRC has warned that such systems “may fail to distinguish combatants from civilians, assess proportionality, or recognize surrender or injury,” especially when used outside tested conditions.
Beyond compliance, there is the question of accountability. When a machine selects and engages, attributing criminal responsibility for war crimes becomes legally and morally fraught. Blame can be diffused among commanders who authorized deployment, operators who pressed “go,” developers who wrote the code, and manufacturers who sold the system. The result, advocates warn, is an accountability gap that undermines deterrence and justice for victims.
Underpinning both legal and accountability concerns is a deeper ethical claim: that delegating life?and?death decisions to algorithms crosses a moral red line and erodes human dignity. This is not merely philosophical; it shapes public legitimacy for the use of force. As Guterres and Spoljaric framed it, the core issue is “the autonomous targeting of humans by machines,” a practice they argue must be prohibited before it becomes normalized.
Investment, proliferation, and the race no one can win
The humanitarian case is backed up by brutal market realities. Start-ups focused on autonomous weapons had raised more than $146 billion in the first five months of 2026, breaking all previous full-year records and owing to huge profit margins that drive quick deployment. The resulting financial tailwind fuels a technology arms race among states and private corporations eager to incorporate increasingly complex artificial intelligence into targeting, detection, and decision-making processes.
Spreading proliferation issues only worsen the situation. Drones have grown from being available to fewer than 10 countries in 2010 to more than 40 countries and 65 non-state actors armed groups at present. Autonomous software, unlike bulky machinery, can be easily replicated and proliferated at low cost, raising the fear of potential access by smaller countries, criminal organizations, or even individuals. According to the ICRC and UN, without regulation the whole world may find itself in a future dominated by autonomous weapons, with advanced systems being developed first by powerful countries and inferior versions by others.
This dynamic feeds what advocates describe as “algorithmic coloniality”: a global order in which a handful of technology?rich states and corporations set the norms and capabilities that everyone else must live with, while the humanitarian costs are borne disproportionately by populations in conflict zones. The UN General Assembly resolutions have explicitly warned of proliferation to unauthorized recipients and non?state actors, underscoring the global security stakes.
