Over the past two decades<\/a>, artificial intelligence (AI) in the military domain, which includes a reliance on algorithmic decision-support systems<\/a> and predictive pattern-of-life targeting<\/a>, has dramatically altered how war is experienced<\/a>, and how crucial parts of International Humanitarian Law (IHL) \u2013 such as civilian status \u2013 are interpreted.<\/p>\n\n\n\n In early September 2026, the final meeting<\/a> of the United Nations group of governmental experts (GGE) on emerging technologies in the area of Lethal Autonomous Weapons Systems<\/a> concluded a final text of \u201celements for an instrument\u201d, adopted by states, available to read online<\/a><\/a>. The text will be presented to delegations convening in Geneva, the Swiss capital of diplomacy, at the seventh review conference<\/a> of the Convention on Certain Conventional Weapons<\/a> \u2013 commonly referred to as the \u2018CCW\u2019 \u2013 which will be held this month from 16 to 20th November 2026.<\/p>\n\n\n\n A reading of the text<\/a> \u2013 often referred to as \u201cregulating killer robots<\/a>\u201d \u2013 generates a disturbing disconnect from the reality of today\u2019s wars and raises questions about whether a treaty in this forum could ever serve to restrain autonomous or artificial intelligence-assisted uses of force.<\/p>\n\n\n\n The 1980 CCW<\/a> is an international treaty banning or restricting the use of weapons considered to cause unnecessary or unjustifiable suffering to combatants and indiscriminate harm to civilians, in both international and non-international armed<\/a> conflicts. It operates through five annexed protocols<\/a>, designed to be added to, which address specific categories of harmful weapons:<\/p>\n\n\n\n Protocol I (1980) on Non-Detectable Fragments<\/p>\n\n\n\n Protocol II (1980, amended 1996) on Mines, Booby-Traps and Other Devices<\/p>\n\n\n\n Protocol III (1980) on Incendiary Weapons<\/p>\n\n\n\n Protocol IV (1995) on Blinding Laser Weapons<\/p>\n\n\n\n Protocol V (2003) on Explosive Remnants of War<\/p>\n\n\n\n Could there be a sixth protocol on Lethal Autonomous Weapons Systems (also known by the acronym LAWS)? Possibly. The CCW was designed to be a flexible convention, so new weapons protocols could be added to it over time. Under article 8 of the CCW, any state that has ratified the treaty has the power to propose an additional protocol relating to categories of conventional weapons not yet covered, and a conference to negotiate a protocol can proceed if a majority of states agree. Or, states can collectively decide on new protocol negotiations at a meeting.<\/p>\n\n\n\n Because the CCW<\/a> operates by consensus, a decision can only be made if no member state actively objects to it. This enables broad participation on the one hand, but on the other hand means that the CCW requires the passive or explicit agreement of every single participating state, which has implications for how text is negotiated \u2013 vetoed and blocked, or compromised and diluted. Some states embrace consensus and others instrumentalise it, diluting the collective will and ambition, as Yvette Issar from the Quaker UN Office in Geneva, has reflected on<\/a>.<\/p>\n\n\n\n An example of how past deadlock has forced frustrated states to negotiate binding bans outside UN structures through independent processes is the 2008 Convention on Cluster Munitions<\/a>. After states faced deadlock at the 2006 CCW Third Review Conference, with military powers and producers such as the United States, Russia and China blocking and stalling proposals, the states that were committed to introducing a ban on cluster munitions<\/a> chose instead to engage in a separate process initiated by Norway, referred to as the \u201cOslo process\u201d. It did not require universal consensus and so it moved incredibly fast. Over 100 states successfully negotiated and adopted only two years after the Convention on Cluster Munitions<\/a> on 30 May 2008 in Dublin. <\/p>\n\n\n\n According to Stop Killer Robots\u2019 Automated Research tracking data<\/a> at the beginning of October, 76 states have publicly expressed their support for moving forward to negotiations<\/a> on a LAWS instrument \u2013 a figure that has since risen to 84<\/a>. Furthermore, 130 states<\/a> have expressed support for some form of legally binding instrument and, in October 2026, 163 states voted in favour<\/a> of pressing ahead with this process at the UN General Assembly<\/a>, building on previous UNGA votes<\/a>.<\/p>\n\n\n\n The UK has agreed with UN resolutions on the risks of autonomous weapons, but it has resisted backing a legally binding international treaty<\/a>. However, in a Stop Killer Robots UK panel event<\/a> on the fringes of the Labour Party Conference in late September, panellists concluded that ahead of the November review conference, the UK should support a binding regulation<\/a>.<\/p>\n\n\n\n \u201cThe reason we need an actual legally binding instrument, which effectively specifies how it applies, is because if we are not clear as to what is legal and what is not in the use of these technologies, it's up to interpretation\u201d, said Alexi Drew<\/a>, a former Technology Policy Adviser at the International Committee of the Red Cross. \u201cWe\u2019ve seen what interpretation does in this current climate of armed conflict between peers \u2013 it expands\u201d, she added<\/a>.<\/p>\n\n\n\n The text which will be presented this month, is a political compromise which establishes a definition<\/a> of lethal autonomous weapon systems within the scope and application of the CCW.<\/p>\n\n\n\n \u201cA lethal autonomous weapon system can be characterized as a combination of one or more weapons and technological components operating together, that is capable of identifying<\/a>, selecting, and engaging a target in a way that results in loss of life, without intervention by a human being operating the system\u201d, reads paragraph 25<\/a> of the draft.<\/p>\n\n\n\n The applicable legal framework of the text is restricted to IHL, excluding International Human Rights Law<\/a> despite its obvious and related connections to the issue \u2013 including policing, border control and security. Weapons which cause unnecessary suffering<\/a> or that cannot be used in a way that allows discrimination between combatants and civilians are prohibited. The CCW seeks to ensure that there maintains a distinction between these categories and that civilians are protected<\/a>, providing additional safeguards to regulate the means and methods of warfare adopted by parties to armed conflict. Other inclusions in the text note that control must rest with humans, not machines \u2013 and human judgement must be exercised to uphold IHL<\/a>.<\/p>\n\n\n\n A closer examination shows how the final closed-door negotiations changed an earlier draft version of the rolling text<\/a>, dated 5 June 2026. This draft text included stronger language on requirements for \u201chuman control\u201d and \u201cethical consideration\u201d, however by the final version<\/a>, these terms were replaced with more permissive phrasing of \u201ccontrol and human judgement\u2019, reflecting the type of compromised language that is preferred by states including the UK<\/a>, and allowing for human control to be exercised \u201cdirectly or indirectly\u201d.<\/p>\n\n\n\n Similarly, any outright prohibitions were stripped from the text entirely. The previously suggested mention that \u201cStates shall refrain from manufacturing, otherwise acquiring, stockpiling, or transferring LAWS that cannot be used in compliance with IHL\u201d was removed from the final text. So too were the lines requesting \u201cthat LAWS can be deactivated or neutralized in a timely manner\u201d and \u201censuring that LAWS, in their identification, selection, and engagement functions, are adequately predictable and reliable in operation, and that their functioning and effects are traceable and explainable to those responsible for their development and use\u201d, removing reference to predictability, reliability, traceability and explainability.<\/p>\n\n\n\n Rather than pushing for a prohibition, the text\u2019s permissive guidelines leave open doors through which automated warfare will continue to strain<\/a>, and break, established legal protections<\/a>.<\/p>\n\n\n\n The final text mentions civilian protection, but how those provisions are framed is problematic \u2013 leaning into procedural risk mitigation, requiring states to test systems, consider risks and implement measures against unintended or automation bias<\/a> \u2013 rather than considering previously harmful bias<\/a> and instead of establishing structural legal prohibitions<\/a> that can be understood, and adhered to or challenged, after an attack<\/a>.<\/p>\n\n\n\n The text requires states to implement measures to intervene or deactivate these weapons, but adds a glaring loophole that they must only do so \u201cto the extent possible or feasible\u201d. This caveat enables militaries to deploy systems they know they cannot effectively recall.<\/p>\n\n\n\n Even considering LAWS as a separate issue, which of course it is not<\/a>, another concern is how the text establishes itself within existing IHL<\/a> \u2013 emphasising risk management, human command and control and lifecycle accountability<\/a> rather than establishing structural legal prohibitions. The text insists that \u201caccountability cannot be transferred to machines,\u201d however by decoupling LAWS from broader military AI and focusing on technical fixes for fundamental threats such as bias<\/a>, whether deliberate or not, this fragmentation obscures the ways<\/a> in which harm is conducted.<\/p>\n\n\n\n As conflicts increasingly digitise and civilians interact<\/a> with crowdsourced intelligence<\/a>, including mobile applications<\/a>, open-source cyber operations<\/a> and commercially available drones<\/a> and other dual use objects and technologies<\/a>, the temporal and behavioural boundaries of a key IHL issue \u2013 the direct participation in hostilities<\/a> of civilians \u2013 is also at risk of being dangerously undermined<\/a>.<\/p>\n\n\n\n Civilians must continue to be protected by a strict \u2018rule of doubt\u2019, which means that in cases of doubt, a person is presumed to be a civilian.<\/p>\n\n\n\n Automated targeting webs and compressed Intelligence Surveillance and Reconnaissance (ISR) cycles risk conflating survival-driven civilian behaviour with hostile intent, flattening distinction between protected populations and targets, leaving gaping accountability gaps when civilians are slaughtered<\/a>. Algorithmic warfare relies on predictive algorithms to generate target lists, and military AI accelerates this legal abstraction<\/a> with civilians being implicitly forced to prove their non-combatant status through behaving in ways that algorithms can recognise.<\/p>\n\n\n\n On that, the International Committee of the Red Cross has called<\/a> for a prohibition on unpredictable systems<\/a> and anti-personnel LAWS<\/a>, among other recommendations, entirely taking targeting people off the table<\/a>.<\/p>\n\n\n\n Momentum must continue to grow outside Geneva\u2019s CCW process, whatever happens this month. Elizabeth Minor, Head of Policy at Stop Killer Robots, points out that, \u201cto seriously and holistically address the humanitarian, ethical, and legal threats from AI and automation in the use of force, states will need to do much more than make baseline agreements on how IHL applies in war to these new technologies. When it comes to autonomous weapons systems, systems that autonomously target people and those that can\u2019t be meaningfully controlled must be clearly prohibited<\/a> in international law\u201d.<\/p>\n\n\n\n \u201cMost urgently<\/a>\u201d, she adds, \u201cwe are demanding that states put an immediate moratorium on using any AI system, including generative AI, as the basis for targeting decisions, given the shocking reports we have been seeing of conflict parties using AI to inform who lives and dies without meaningful deliberation on the outputs produced by systems. This dehumanising and reckless automation of violence is unacceptable and must stop now: life and death decisions should never be left to machines.\u201d<\/p>\n\n\n\n When states convene in Geneva this month, delegations must display the political courage to demand binding legal limits<\/a>, backed by the widest possible outreach, education and engagement, and press forward with a mandate to negotiate a legal protocol within one year<\/a>.<\/p>\n\n\n CHECKING STATES\u2019 POSITIONS<\/strong><\/p>\n To check a state's position, Justice Info\u2019s readers might see the Automated Decision Research<\/a> States Position Monitor<\/a>. They can see also the list of countries<\/a> that have said specifically that they are ready to move to negotiations based on the work done at the CCW Group of Governmental Experts. You can check if your state is party to the CCW here<\/a>. And, see the regional briefing papers<\/a>\u00a0for more information on collective and regional statements and positions, available online<\/a> in English<\/a>, Arabic<\/a>, French<\/a>, and Spanish<\/a>.<\/p>\n<\/div>\n\nPossibly a sixth protocol to the CCW?<\/h2>\n\n\n\n
156 states in favour of moving forward<\/h2>\n\n\n\n
![\"Images](\"https:\/\/www.justiceinfo.net\/wp-content\/uploads\/Austria_Military-Autonomous-Weapons-Regulation-Conference_@Joe-Klamar-AFP.jpg\")
\u201cLoss of life without intervention of a human being\u201d<\/h2>\n\n\n\n
Prohibitions stripped from the initial text<\/h2>\n\n\n\n
What does this mean for civilian protection?<\/h2>\n\n\n\n
\u201cLife and death decisions should never be left to machines\u201d<\/h2>\n\n\n\n
![\"Military](\"https:\/\/www.justiceinfo.net\/wp-content\/uploads\/military-AI_@shutterstock-540x360.jpg\")
<\/div><\/a>\r\n\t\t\t![\"Alannah](\"https:\/\/www.justiceinfo.net\/wp-content\/uploads\/Alannah-Travers.jpg\")
ALANNAH TRAVERS<\/strong><\/p>\n