Every afternoon at 17:00 during the Vietnam war, U.S. officials gathered the Saigon press corps to explain how well the war was going. Maps, operational statistics and enemy body counts were neatly displayed — all to back up the same reassuring story: the strategy was working and victory was getting closer.
In reality, of course, the war was going disastrously bad. The journalists — who had seen the war with their own eyes — knew that. The U.S. officials knew that the journalists knew — and the journalists knew that too.
Yet everyone returned every afternoon and resumed the absurd performance. What the Saigon press corps would later dub “the five o’clock follies” stands as an example of a process that continues to manufacture a narrative of progress long after its participants have stopped believing in the underlying story.
It would have been laughable had the suffering it concealed not been so suffocatingly great.
At exactly 2:57 in the morning of Saturday 5 September, the diplomats still awake in Room III of the Palais des Nations in Geneva erupted in applause. From his seat beneath L’harmonie — an eight-meter-wide mural depicting five women in a boat — Ambassador Robert in den Bosch of the Netherlands had just gavelled the final meeting of the Group of Governmental Experts on lethal autonomous weapon systems to a close.
Much would later be made about the early-morning time stamp — as if keeping a meeting going into the wee hours is, in itself, a measure of success.
But for the Dutch Ambassador, one thing appears to have mattered above all: none of the 128 parties to the Convention on Certain Conventional Weapons (CCW) had objected when he proposed, after three years of discussions, that they adopt the report he had so meticulously crafted.
In this “difficult geopolitical situation,” he wrote the next day, the group had managed to fulfil its mandate and, through “seriousness, expertise and commitment”, contributed “meaningfully to the collective effort to uphold humanity in armed conflict.”
Just as in L’harmonie — painted in 1936, ironically, while Europe was drifting towards another world war — the Ambassador had kept the meeting afloat, even when its participants had appeared preoccupied with things other than actually getting somewhere.
And at face value, it really did sound like a remarkable achievement: In a world in which arms build-ups, bombs and threats of conquest have reemerged as the grammar of international politics, bitter rivals had, it seemed, found common ground.
Envoys from countries as hostile to each other as Russia and Ukraine, India and Pakistan, the United States and China, Israel and the State of Palestine had agreed — or at least not found sufficient cause to object — to a text addressing one of the most contentious questions in modern warfare: how AI may be used to identify, select and attack targets without human intervention.
But the price of the veneer of harmony had been high. Over the past two years, the Ambassador had released at least eight successive versions of a “rolling text” outlining “a set of elements of an instrument” on autonomous weapon systems.
His first text, released in the summer of 2024, had — beyond a “working characterisation” and a description of how international humanitarian law applies to such weapon systems — drawn a set of clear red lines: the use of autonomous systems “whose effects cannot be anticipated or controlled” and those that “operate without appropriate control or human judgement” should be prohibited.
The Ambassador had also suggested that states put in place common safeguards and specific operational limits on how autonomous weapons must be used: any such system “must be adequately predictable, reliable, traceable and explainable”, he wrote.
Two years and more than seven rewrites later, little of that remained. By 2:57 on Saturday 5 September, after a final week of discussions in Geneva, most of the red lines had been erased; the few that remained had faded to a pale pink.
True, a version of the Ambassador’s “working characterization” had survived the strikeout job. But almost everything that could have provided real normative guidance had been pared away. The prohibitions were reduced to a generic restatement of existing law; the safeguards and operational limits had been downgraded to a menu of voluntary measures.
In the end, even the sole reference to “ethics” was negotiated out of the text. The United States, Russia and other militarised states had succeeded, it seemed, in blocking anything that would have placed limits on their military operations.
“We didn’t have the tenacity to continue arguing with them, because if they said ‘we don’t agree with that’, the Chair would just strike that provision. That’s how it was progressing,” a negotiator participating in the closed-door discussions recalled.
The problem had not been lack of support. 130 countries stand in favour of new legal rules on autonomous weapons. Earlier in the week, more than 70 of them had made it clear that a version of the rolling text — one that preserved the red lines from the Ambassador’s initial version — should “serve as a basis for negotiations” on a new instrument.
Neither had it been lack of public backing. In May, a cross-regional Lex International/Ipsos poll found that majorities in every country surveyed — including 79 per cent of Chinese and 63 per cent of Ukrainians — support a ban on the use of AI to target and attack people.
The problem had been that the procedures governing the talks were stacked against the rule-making majority. Three years earlier, states had agreed that the group would “conduct its work and adopt its report by consensus.” No doubt a nice aspiration, but consensus had come to be treated as an absolute requirement — effectively giving every country a veto. That, of course, created a structural bias towards inaction: those seeking to block new rules needed only to say no.
“The law of the least ambitious programme”, the late Norwegian political scientist Arild Underdal called it: consensus-based negotiations tend to converge on the preferences of the most reluctant participant.
The rule-making majority knew that they would ultimately be unable to defend their red lines. “There is a consistent pattern of which countries can block, and which countries cannot,” one negotiator explained in an interview. “Every delegation has red lines. But some red lines are more sacrosanct than others.”
Observers, barred from the closed-door proceedings, had scrambled to put the pro-treaty bloc on notice. On Friday 4 September, as representatives of the blocking minority chipped away at the majority’s red lines, the text at one point became so weak as to lead the International Committee of the Red Cross (ICRC) to recommend — according to several negotiators — that pro-treaty delegations simply reject it.
A few weeks before the meeting, ICRC President Mirjana Spoljaric had called on states to “urgently adopt a legally binding instrument to regulate autonomous weapon systems with clear prohibitions and restrictions”, together with UN Secretary-General Antonio Guterres.
“States have a responsibility to those already suffering the consequences of today’s conflicts, and to those who will inherit the world that is now being shaped,” the two leaders pleaded.
But pro-treaty negotiators did not have instructions from their capitals to act on the recommendation. “Nobody really considered, in all seriousness, the prospect of blocking the report in its entirety”, a pro-treaty negotiator recalled.
By contrast, the blocking minority did not even have to threaten to break consensus — it was already assumed that they would block and accept no outcome if their red lines were crossed. “Everybody knew, including the Chair,” a negotiator said.
Without instructions to withhold consensus themselves, the delegates trying to preserve a minimum of ambition were ultimately unable to stop the drift towards the lowest common denominator.
Shortly after 2:00 on Saturday 5 September, a final — and substantially weakened — version of the report was circulated. Delegates had barely ten minutes to read it before the meeting resumed under significant time pressure. One negotiator recalled the Chair warning that the room had to be vacated by 3:00, when the lights would go out.
Pro-treaty negotiators knew, at that point, that it was over. “We had embraced our fate with a disheartened attitude at the eleventh hour”, one of them said. “We were getting consensus because we were not objecting anymore”.
In the days following the adoption of the final report, ambitious negotiators did their best to put a positive spin on the outcome. “It was an extremely difficult negotiation”, Claudio Medeiros Leopoldino, Brazil’s lead negotiator, wrote on LinkedIn. Yet there is “widespread hope” that the CCW Review Conference in November will decide to begin negotiations on a new instrument, he added. Despite “painful language compromises,” Reto Wollenmann of Switzerland concluded, “we have a logic of a future instrument agreed”. The report marks a “significant step forward towards an international instrument governing autonomous weapons systems”, Steven Murphy of Ireland followed up.
Hope has a place in multilateral diplomacy. Yet none of these accounts provides a credible political narrative of how, precisely, treaty negotiations will come about. The adopted report recommends that the CCW Review Conference decide “by consensus on the possible modalities of the future work of the Group” — locking, it would seem, the question of whether to start treaty negotiations into the exact same veto-driven trap that left the rule-making majority so painfully disheartened in the first place.
As the Peruvian civil society observer Gisela Luján Andrade noted: “within the CCW it will be difficult to achieve an instrument that places the spirit of civilian defense and protection at its core”. Indeed, as long as Russia, the United States and other militarised states remain dead set against any new legal limits on their military operations, the rules needed to prevent machines from killing people in war will not emerge from a consensus-bound arena like the CCW.
The rule-making majority knows this; the blocking minority knows that the rule-making majority knows this — and the rule-making majority knows that too.
It is now thirteen years since the parties to the CCW first decided to “discuss the questions related to emerging technologies in the area of lethal autonomous weapon systems.” The fate of the Dutch Ambassador’s rolling text shows that expert-level discussions have hit the end of the road. To respond to the strong public opposition to the use of AI to target and attack people, high-level political leadership is urgently needed.
At the UN General Assembly last week, Norwegian Prime Minister Jonas Gahr Støre and Finnish President Alexander Stubb launched a new cross-regional initiative to keep AI under “human direction, oversight and control”. Stubb listed “killer robots using AI for lethal attacks“ among the main threats the initiative — now endorsed by leaders from 28 countries — aims to counter.
Talk, however, is relatively cheap in the UN these days. Delivering will require a willingness to move ahead and develop the required legal norms, even when the blocking minority objects. The day after Støre and Stubb launched their initiative, Donald Trump countered by announcing that the United States “totally rejects any attempt to construct a globalist scheme of control” for AI.
Expecting the United States and other states opposed to new rules on autonomous weapons to agree to meaningful negotiations would be naïve. Waiting for their consent before putting in place the rules people demand would be folly.
“We must not allow the lack of interest in some quarters to prevent small and medium-sized countries from initiating a process to fulfil our humanitarian obligations,” Støre wrote in 2006. Then, as now, a small group of states stood in the way of new legal rules in the CCW. Then, as now, the only way out of the impasse is for a coalition of committed countries to act.
Eighteen months after Støre issued that challenge 107 states adopted the Convention on Cluster Munitions in Dublin, creating “the most impactful of all legal instruments we’ve seen on the weapon side, hands down,” as the ICRC’s Erik Tollefsen said recently.
In the world outside the UN conference rooms, tech companies and their powerful military clients are developing and deploying new means of killing at breakneck speed. In August, The New York Times reported that a Russian AI-guided drone had killed the 19 year old student Tetiana Bubynets and two other civilians in eastern Ukraine.
Against that backdrop, one may hope that Norway and other committed states have not forgotten what an effective treaty-making process requires — and that the point of disarmament diplomacy is not to agree at any cost, but to protect people from war.
Write for Spoiler Alert
Spoiler Alert is inviting pitches from journalists interested in exploring under-reported and overlooked stories behind the laws of war and peace. Journalists with relevant ideas or materials are encouraged to contact Magnus Løvold at magnus@lexinternational.org