There is no way each and every one of the 1,000 targets a day are thoroughly assessed.
At a military conference in London, an Israeli arms executive revealed that Israel's Tzayad digital command system identified 850,000 potential targets across the wars in Gaza and Lebanon — roughly 1,000 per day — a figure that experts say exceeds any realistic capacity for human oversight of civilian harm. The disclosure, made beside NATO's second-ranking officer, places in sharp relief a tension as old as warfare itself: the gap between the speed at which modern systems can act and the moral deliberation that just conduct demands. As artificial intelligence compresses the window between identification and strike from nearly an hour to mere minutes, the question of who bears responsibility for each death becomes harder to answer — and easier to defer.
- An Israeli general quietly presented a staggering number — 850,000 targets detected in real time — to a room of NATO officers, reframing the scale of two simultaneous wars in a single slide.
- Experts warn that 1,000 daily targets cannot receive meaningful collateral damage review; AI systems like Lavender and Hasbora reduced human review of individual targets to as little as 20 seconds each.
- Elbit Systems disputed the word 'targets' in its own slide, calling the figure 'aggregated system activity' — a semantic retreat that former Pentagon targeting adviser Wes Bryant flatly rejected.
- Over 71,000 Palestinians were killed in Gaza during the period covered by the data, more than half of them women, children, and elderly — deaths that occurred within the same operational window the system was running.
- NATO militaries are watching closely, aware that the counterinsurgency-era practice of deliberate, legally vetted targeting has given way to a tempo that may have structurally outpaced human accountability.
At a Royal United Services Institute conference in London last week, Israeli reserve Major General Miki Edelstein — also executive vice-president of Elbit Systems, Israel's largest arms manufacturer — presented a figure that stopped military experts in their tracks: 850,000 targets detected in real time by Israel's Tzayad digital command system across the Gaza and Lebanon wars between October 2023 and the end of 2025. Seated beside him was NATO's second-ranking officer. The number works out to roughly 1,000 potential targets flagged each day — people, vehicles, structures. Edelstein noted that Tzayad had compressed the time from target identification to strike from 40–50 minutes down to one to seven minutes, with a human approving each strike. He called that oversight "the right thing to do."
Wes Bryant, a former senior Pentagon targeting adviser, heard the presentation and calculated that 850,000 targets across Gaza — a territory of 2.2 million people and 300,000 buildings — meant the IDF had at some point aimed at more than half of the strip's entire population and infrastructure. Cambridge researcher Sophia Goodfriend argued that properly vetting 1,000 targets a day was simply impossible without delegating most of the work to automated systems, raising acute questions about legal accountability when strikes go wrong.
Elbit pushed back, claiming the figure referred to "aggregated system activity," not targets — despite the word appearing on the slide itself. Bryant was unconvinced. "There is no way," he said, "that each and every one of the 1,000 targets a day are thoroughly and effectively characterised in terms of collateral damage analysis."
Tzayad operates alongside two other AI systems: Lavender, which once flagged 37,000 people as potential Hamas affiliates based on algorithmic pattern-matching, and Hasbora, which could generate 100 strike recommendations per day. One intelligence officer said Lavender targets received roughly 20 seconds of human review each. Early in the Gaza war, officers reportedly approved strikes that could kill 15 to 20 civilians if the target was a low-ranking militant.
The World Health Organization counted 71,269 Palestinians killed in Gaza through the end of 2025 — more than half of them women, children, and elderly. In Lebanon, 3,961 died during the autumn 2024 war. Both tolls fall within the operational window Edelstein described. Across NATO, commanders acknowledge that modern warfare now moves faster than the deliberate legal frameworks built during the slower campaigns in Afghanistan and Iraq — and the central unresolved question is whether any human institution can still keep pace with the systems it has built.
At a military conference in London last week, an Israeli general stood before NATO's second-ranking commander and a room full of uniformed officers to present a number that would reshape how we understand modern warfare: 850,000. That was the total count of targets detected in real time by Israel's Tzayad digital command system across the wars in Gaza and Lebanon between October 7, 2023, and the end of 2025. The figure works out to roughly 1,000 potential targets identified each day—people, vehicles, structures, anything the system flagged as a possible enemy position.
Miki Edelstein, a major general in the Israeli reserves and executive vice-president of Elbit Systems, the country's largest arms manufacturer, presented the data at a Royal United Services Institute conference. He was not listed on the agenda by name until the session began. Sitting beside him was Air Chief Marshal Sir Johnny Stringer, NATO's second-most senior military officer. The slide Edelstein showed described what the Israeli military calls "high-tempo operations," powered by more than 20,000 IDF battle plans and those 850,000 real-time intelligence targets. He explained that the Tzayad system helped compress the time between identifying a target and striking it—from 40 to 50 minutes down to between one and seven minutes. A human decision-maker, he noted, still had to approve each strike. It was, he said, "the right thing to do."
Wes Bryant, a former senior targeting adviser at the Pentagon who specialized in assessing civilian harm, heard the presentation and felt alarm. Gaza had a population of 2.2 million people and 300,000 buildings before October 2023. If the IDF had targeted 850,000 objects across the territory, Bryant calculated, the military had at some point or another aimed at more than half the entire population and infrastructure of the strip. He said the number was deeply concerning. Sophia Goodfriend, a researcher at Cambridge University studying artificial intelligence in warfare, went further: it would be nearly impossible for any military to properly vet 1,000 targets a day without handing much of the verification work to automated systems. That raised hard questions about who was actually accountable when a strike went wrong, and whether human judgment was being squeezed out of the process.
Elbit disputed the interpretation. When the Guardian asked about the 850,000 figure, a company spokesperson said it did not refer to targets at all, but rather to "aggregated system activity and operational data." The slide itself had said "targets." The spokesperson clarified that the numbers showed the volume of information the Israeli military was processing, not the number of enemy positions or actual strikes. Bryant was unmoved. Even assessing 50 targets a day for collateral damage and legal risk was difficult, he said. Assessing 1,000 was not realistic. "I will say, definitively, that there is no way each and every one of the 1,000 targets a day—let alone 850,000 targets in aggregate—are thoroughly and effectively characterised in terms of collateral damage analysis and assessed risk to civilian populations."
The Tzayad system is only one piece of Israel's targeting infrastructure. The military also uses two other AI-powered databases called Lavender and Hasbora. Lavender once identified 37,000 people as potential Hamas targets based on algorithmic assessment of their connections. Hasbora could generate 100 targets a day and recommended buildings to strike. An Israeli intelligence officer said targets flagged by Lavender were reviewed by a human for about 20 seconds each because the system produced so many. Two other intelligence officers said that during the early stages of the Gaza war, it was permissible to kill 15 or 20 civilians in a single airstrike if the target was a low-ranking militant.
The human toll is documented. The World Health Organization counted 71,269 Palestinians killed in Gaza through the end of 2025, the primary theater during the period Edelstein cited. More than half were children, women, and elderly people. In Lebanon, 3,961 people died during the autumn 2024 war, about a quarter of them women and children. These deaths occurred within the operational window when the Tzayad system was identifying its 850,000 targets. Military leaders across NATO countries have noted that modern wars are being fought at a faster pace than the counterinsurgency campaigns in Afghanistan and Iraq, where there was more time to consider whether a target was legal. The question now is whether speed and scale have outpaced the ability of any human system—or any human being—to ensure that the targets being struck are actually what the algorithms say they are.
Citações Notáveis
I will say, definitively, that there is no way each and every one of the 1,000 targets a day—let alone 850,000 targets in aggregate—are thoroughly and effectively characterised in terms of collateral damage analysis and assessed risk to civilian populations.— Wes Bryant, former Pentagon targeting adviser
Any military would struggle to do so without outsourcing verification to other automated systems, which raises questions of accountability and concern about shrinking amounts of human oversight.— Sophia Goodfriend, Cambridge University researcher on AI in warfare
A Conversa do Hearth Outra perspectiva sobre a história
When Elbit says 850,000 targets, what exactly are they counting?
The company later claimed it was just system activity—data flowing through the network. But the slide said targets. And if you take the number at face value, it means the military identified something to shoot at roughly every 90 seconds for two years straight.
Is that even possible to do safely? To check each one?
No. A former Pentagon targeting officer said you'd need to assess collateral damage and legal risk for each strike. He said even 50 a day is hard. A thousand is impossible without letting machines do most of the work.
So machines are deciding what's a target?
Not deciding—but verifying. Lavender flagged 37,000 people as potential targets. A human looked at each one for about 20 seconds. That's not really vetting. That's rubber-stamping.
Why does speed matter so much?
Because the faster you can go from detection to strike, the more you can hit before the target moves or hides. Tzayad cut the time from 50 minutes to seven minutes. That's a massive advantage in combat. But it also means less time to ask whether you're hitting the right thing.
And the people who died—were they all targets?
We don't know. Over 71,000 Palestinians were killed in Gaza during this period. More than half were children, women, and elderly. The system was designed to find enemies. Whether it did is the question nobody can answer anymore.