This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Investigations report that the Israel Defense Forces utilized AI systems, specifically 'Lavender' and 'The Gospel', to generate target lists of suspected militants in Gaza. These systems, combined with tracking software like 'Where's Daddy?', were used to authorize airstrikes on residential homes, resulting in significant civilian casualties. Sources indicate that human oversight was minimal, with officers often acting as a 'rubber stamp' for AI-generated targets.
The AI system known as "Lavender" was reportedly used by the Israel Defense Forces (IDF) to assist in identifying individuals in Gaza for targeting, while a related system, "The Gospel," was reportedly used to help select and prioritize physical strike targets. Investigations reportedly suggest that these systems operated with limited human review and may have contributed to strikes associated with high civilian casualties. The extent of automation with human oversight, as well as the accuracy of these targeting processes, reportedly remain subjects of dispute.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to develop cyber weapons (e.g., by coding cheaper, more effective malware), develop new or enhance existing weapons (e.g., Lethal Autonomous Weapons or chemical, biological, radiological, nuclear, and high-yield explosives), or use weapons to cause mass harm.
Also classified as
Human operators delegated critical targeting decisions to the AI, spending only 20 seconds to rubber-stamp recommendations.
The Lavender system had an acknowledged 10 percent error rate, frequently misidentifying civilians, civil defense workers, and police as militants.
The use of the Lavender and Where's Daddy? AI systems to systematically target and execute military strikes causing mass civilian casualties constitutes the use of AI weapons to cause mass harm.
AI system
Due to a decision or action made by an AI system
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by the deployment of the Lavender and Where's Daddy? AI systems, which were intentionally used post-deployment to generate and track lethal military targets.
Added categories peak at 4: Severe (Psychological Harm, Indirect)
The deployment of Lavender, The Gospel, and Where's Daddy? represents the first documented case of mass-scale AI systems directing lethal military targeting in active warfare. By reducing human oversight to a rubber stamp, the system caused unprecedented civilian casualties and infrastructure damage, establishing a dangerous global precedent for automated warfare and violating international humanitarian law.
The incident involves multiple specialized AI systems working in coordination: Lavender generates target lists, Where's Daddy? tracks targets' real-time locations to their homes, and The Gospel identifies structural targets. This multi-step pipeline of distinct AI models orchestrates the automated targeting process.
Unacceptable Risk: The report describes an AI system that performs mass surveillance and algorithmic profiling of an entire population, assigning a threat score of 1 to 100 to individuals, which is highly analogous to prohibited 'social scoring by governments' under Risk Level 1.
The primary function of the Lavender and The Gospel systems is to identify specific individuals and structures as military targets for airstrikes.
The reports are highly detailed, drawing on extensive investigative journalism from multiple reputable outlets (+972, Local Call, The Guardian, NYT) and first-hand testimonies from six anonymous Israeli intelligence officers who operated the systems. The core facts regarding the existence of Lavender, Where's Daddy?, and The Gospel, their error rates, and the minimal human oversight are consistent across all sources.
The IDF interprets these systems purely as auxiliary databases meant to assist human analysts, rather than automated decision-making systems that direct lethal strikes.
Technology
High AI Error Rate
The Lavender targeting algorithm has an acknowledged 10% error rate.
Algorithmic Tracking Faults
Systems linked targets to private homes where families were present.
Loose Feature Matching
Model flagged targets based on weak features like shared WhatsApp groups.
Data Inputs
Flawed Training Data
Training data included civil defense and police officers as militants.
Mass Surveillance Data
Opaque profiling of 2.3 million people using mass surveillance inputs.
Outdated Phone Linking
System failed to track real-time phone sharing or device transfers.
Human Factors
Automation Bias
Operators treated machine outputs as if they were human decisions.
Minimal Human Review
Analysts spent only 20 seconds verifying targets, checking only if male.
Cognitive Pressure
Commanders pressured intelligence personnel to bring more targets quickly.
Process and Methods
Automated Target Generation
Shift to automatic target generation bypassed traditional validation steps.
Use of Dumb Bombs
Army used unguided munitions on junior targets to save expensive smart bombs.
Out-of-Date Verification
Target presence at home was not verified in real-time before airstrikes.
Management
Prioritizing Target Quantity
High-level decisions prioritized mass target production over accuracy.
Loosened Collateral Thresholds
Management authorized high civilian casualty limits per target.
Lack of System Accountability
Command structures deflected accountability to algorithms.
Regulatory Environment
Lack of AI Weapon Regulation
Absence of international legal frameworks governing autonomous targeting.
Permissive Rules of Engagement
Loosening of international law rules regarding collateral damage thresholds.
Lack of Independent Oversight
No external monitoring or accountability mechanisms for military AI use.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
