Israeli Military AI System Processed 850,000 Targets in Gaza and Lebanon, Defense Contractor Reveals
Elbit Systems disclosed that its Tzayad digital command program identified 850,000 real-time targets over two years of conflict, sparking intense debate over the feasibility of human oversight in AI-driven warfare.
By Sofia Matos
- Defense and Operational
- Argue that AI systems are essential for processing vast amounts of battlefield data to protect friendly forces and accurately identify threats in high-tempo operations.
- Targeting and Legal Experts
- Warn that the sheer volume of AI-generated targets makes meaningful human oversight and collateral damage assessment impossible.
- Human Rights Advocates
- Contend that the reliance on automated targeting systems facilitates disproportionate warfare and mass civilian casualties without accountability.
Perspectives this story doesn't cover
- Civilians living in the targeted areas
- Independent algorithmic auditors
Why this matters
The unprecedented scale of AI deployment in active combat zones is fundamentally altering the speed of warfare, raising urgent questions about international humanitarian law and whether human operators can meaningfully assess collateral damage when algorithms generate up to 1,000 targets a day.
A major Israeli defense contractor has revealed that its digital command system identified 850,000 potential targets during the first two years of the conflicts in Gaza and Lebanon, highlighting the unprecedented scale of artificial intelligence in modern warfare. The figure was disclosed during a land warfare conference hosted by the Royal United Services Institute (RUSI) in London.[1][3][4]
Miki Edelstein, an executive vice-president at Elbit Systems and a reservist major general in the Israel Defense Forces (IDF), presented the data to an audience that included senior British and NATO military commanders. A presentation slide detailing the IDF's "high-tempo operations" cited the generation of 850,000 "real-time intel targets" between October 2023 and the end of 2025.[1][4]
The system in question, known as the Tzayad (Hunter) digital army program, is designed to map the battlefield by tracking the positions of friendly units alongside detected enemy personnel, vehicles, and infrastructure. By aggregating vast streams of sensor data, the software aims to provide a comprehensive operational picture for tactical decision-making.[1]
Following the presentation, Elbit Systems sought to clarify the staggering figure. A company spokesperson stated that the 850,000 number reflected "aggregated system activity and operational data" processed by the software across all theaters, rather than a direct count of confirmed enemy targets or actual military strikes.[1]
Despite the clarification, the sheer volume of data—averaging roughly 1,000 potential targets generated per day—has alarmed military analysts and targeting experts. Wes Bryant, a former senior targeting adviser at the US Pentagon, stated that it is definitively impossible for human operators to thoroughly assess collateral damage and civilian risk at such a massive scale.[1][2]
Despite the clarification, the sheer volume of data—averaging roughly 1,000 potential targets generated per day—has alarmed military analysts and targeting experts.
"Even characterizing 50 a day is hard enough," Bryant noted, emphasizing that the legal and ethical requirements of warfare demand rigorous human evaluation that algorithms cannot replace. The reliance on automated systems at this volume raises profound questions about compliance with international humanitarian law.[1][2]
The Tzayad program is part of a broader suite of AI tools deployed by the IDF. Previous reports have detailed the military's use of AI-powered databases, such as "Lavender" and "Habsora" (The Gospel), which were specifically designed to rapidly identify suspected militants and recommend infrastructure for bombardment.[1][5]
During the early stages of the Gaza conflict, systems like Lavender reportedly generated lists of tens of thousands of individuals based on their apparent links to militant groups. Intelligence officers disclosed that human review of these AI-generated targets was sometimes reduced to a mere 20 seconds per recommendation, effectively treating the machine's output as a human decision.[1][5]
Regional media and human rights advocates have pointed to these figures as evidence of a systemic reliance on algorithms that blurs accountability. Critics argue that the automation of the kill chain has directly contributed to the high civilian death tolls and widespread infrastructure destruction observed in densely populated areas of Gaza and southern Lebanon.[2][5]
Conversely, defense contractors argue that AI is an indispensable tool for cutting through the "fog of war." In complex urban environments where adversaries utilize subterranean networks and civilian infrastructure, proponents maintain that rapid data processing is essential to protect friendly forces and identify fleeting threats.[1]
The revelations at the RUSI conference underscore how closely Western militaries are observing the integration of these novel capabilities. With NATO commanders in attendance, the presentation served as both a showcase of technological prowess and a stark preview of future military operations.[1][4]
As artificial intelligence continues to accelerate the pace of conflict, the international community faces an urgent challenge. The debate over the Tzayad system's 850,000 data points illustrates a fundamental tension: balancing the military desire for total battlefield awareness against the legal and moral imperative of human judgment in life-and-death decisions.[1][2][4]
What we don’t know
- Exactly what percentage of the 850,000 identified data points resulted in actual kinetic strikes.
- The specific error rate of the Tzayad system in distinguishing between combatants and civilians.
- How allied Western militaries plan to adapt their own legal frameworks in response to the rapid automation of targeting.
Key points
- Elbit Systems revealed its Tzayad digital army program identified 850,000 real-time targets in Gaza and Lebanon between October 2023 and the end of 2025.
- The system generated an average of 1,000 potential targets per day by aggregating massive streams of battlefield data.
- Elbit later clarified the figure represented aggregated system activity and operational data, not necessarily confirmed enemy targets or strikes.
- Former military targeting experts warned that thoroughly assessing collateral damage for 1,000 targets a day is humanly impossible.
- The revelation adds to growing scrutiny over the IDF's use of AI systems like Lavender and Habsora to automate target generation.
Key terms
- Tzayad (Hunter)
- A digital army command program developed by Elbit Systems that maps friendly and enemy positions in real time.
- Habsora (The Gospel)
- An AI system used by Israeli intelligence to automatically review surveillance data and recommend buildings or infrastructure for bombing.
- Lavender
- An AI-powered database used to identify and list tens of thousands of individuals suspected of militant affiliations.
- Kill Chain
- The military process of identifying a target, dispatching forces, initiating an attack, and destroying the target.
Sources
[1]The GuardianTargeting and Legal ExpertsIsraeli command system identified 850,000 targets in Gaza and Lebanon wars, says supplier
Read on The Guardian →
[2]Al MayadeenHuman Rights Advocates850,000 'targets': Experts alarmed by scale of Israeli AI targeting
Read on Al Mayadeen →
[3]The Times of IsraelDefense and OperationalIsraeli command system identified 850,000 targets in Gaza and Lebanon wars, says supplier
Read on The Times of Israel →
[4]Royal United Services InstituteTargeting and Legal ExpertsLand Warfare Conference 2026: Integrating Novel With Core Capabilities
Read on Royal United Services Institute →
[5]+972 MagazineHuman Rights Advocates‘Lavender’: The AI machine directing Israel’s bombing spree in Gaza
Read on +972 Magazine →
Comments
More in Artificial Intelligence
See all →AI Infrastructure
How FlashAttention Bypasses the GPU Memory Bottleneck to Enable Long-Context AI
5 sources
Open Source Standards
How the Open Source Initiative's 1.0 Definition Excludes the Most Downloaded Open-Weight AI Models
7 sources
Generative Adversarial Networks
How a Generator and a Discriminator Compete to Create Realistic AI Output
8 sources
Machine Learning
How Generative AI Maps the Joint Probability Distribution of Data
5 sources
Every angle. Every day.
Get Artificial Intelligence stories with full source coverage and perspective breakdowns delivered to your inbox.




