'Lavender' and 'The Gospel' AI Systems Reportedly Used in Gaza Targeting Operations with Civilian Harm Allegations

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.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.2 Cyberattacks, weapon development or use, and mass harm

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.

Additional risk subdomains

• 5.2 Loss of human agency and autonomy: Human operators delegated critical targeting decisions to the AI, spending only 20 seconds to rubber-stamp recommendations.
• 7.3 Lack of capability or robustness: The Lavender system had an acknowledged 10 percent error rate, frequently misidentifying civilians, civil defense workers, and police as militants.

Causal factors

• Entity: AI
• Intent: Intentional
• Timing: Post-deployment

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.

EU AI Act risk tier

• Risk tier: 1 Unacceptable

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.

AI system and alleged parties

• AI system: Lavender, The Gospel, Where's Daddy?
• AI purpose: Target Identification; Identification
• Behaviour type: Multi-agent
• Alleged developer: Unit 8200, Israel Defense Forces
• Alleged deployer: Unit 8200, Israel Defense Forces
• Alleged harmed parties: Palestinians, National security and intelligence stakeholders, Gazans

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Catastrophic. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Severe, indirect Severe
• Infrastructure: direct Severe, indirect Severe
• Property: direct Catastrophic, indirect Catastrophic
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Severe, indirect Severe
• Civil rights: direct Severe, indirect Severe
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Severe, indirect Severe
• Psychological: direct Minor, indirect Severe
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Physical

Reported: Yes, the reports explicitly describe extensive physical harm, including tens of thousands of deaths and injuries resulting from airstrikes guided by the AI systems.

Directly caused: The reports state that during the first six weeks of the war, when the military almost completely relied on Lavender, approximately 15,000 Palestinians were killed.

Indirectly caused: Thousands of civilians, including over 13,000 children, were killed as collateral damage in strikes targeting residential homes.

Inferred additional harm: It is likely that tens of thousands of additional individuals suffered severe physical injuries or death under the rubble, with total casualties exceeding 30,000 to 50,000.

Infrastructure

Reported: Yes, the reports explicitly describe catastrophic damage to critical infrastructure in Gaza.

Directly caused: Airstrikes targeting residential homes and buildings identified by Lavender and The Gospel destroyed or damaged nearly 70 percent of Gaza's homes.

Indirectly caused: The destruction led to the collapse of the healthcare system, leaving fewer than one-third of hospitals functioning, and causing catastrophic food insecurity for 1.1 million people.

Inferred additional harm: We infer extensive, long-term damage to water, sanitation, electrical grids, and communication networks across the entire Gaza Strip, requiring billions of dollars and decades to rebuild.

Property

Reported: Yes, the reports explicitly describe massive property damage across Gaza.

Directly caused: The IDF destroyed more than 360,000 buildings and flattened entire residential blocks using unguided dumb bombs on AI-selected targets.

Indirectly caused: More than half of Gaza's buildings and 70 percent of homes were damaged or destroyed as collateral damage from the targeting campaign.

Inferred additional harm: The total value of destroyed residential, commercial, and public property is inferred to be in the billions of dollars, decimating the personal wealth of millions of residents.

Differential treatment

Reported: Yes, the reports describe systematic differential treatment of Palestinians in Gaza through mass surveillance and algorithmic profiling.

Directly caused: The Lavender system profiled and assigned a numerical threat score from 1 to 100 to almost all 2.3 million Palestinian residents of Gaza to identify targets.

Indirectly caused: The military applied highly permissive collateral damage rules to Palestinian targets, allowing up to 20 civilian deaths for low-level targets, which is unprecedented compared to other modern conflicts.

Inferred additional harm: The algorithmic profiling of an entire population based on demographic and communication features likely resulted in systemic discrimination and arbitrary deprivation of life for thousands of innocent individuals.

Civil rights

Reported: Yes, the reports explicitly describe severe violations of human rights and international humanitarian law.

Directly caused: The algorithmic generation of kill lists and subsequent airstrikes violated the fundamental right to life for thousands of individuals, including non-combatants.

Indirectly caused: The systematic bombing of families in their homes violated the Fourth Geneva Convention of 1949, which forbids the indiscriminate killing of non-combatants.

Inferred additional harm: The widespread use of automated targeting with minimal human verification represents a systemic erosion of the right to due process, distinction, and proportionality under international law.

Privacy

Reported: Yes, the reports describe massive privacy violations through the surveillance systems feeding the AI.

Directly caused: The Lavender system analyzed personal data collected on almost all 2.3 million residents of Gaza through a system of mass surveillance, including phone contacts, social media, and location data.

Indirectly caused: The Where's Daddy? system tracked individuals' real-time locations and entry into their private family homes to facilitate airstrikes.

Inferred additional harm: The continuous, non-consensual harvesting of deeply personal communication and location data of millions of civilians represents an unprecedented scale of privacy intrusion.

Psychological

Reported: Yes, the reports describe severe psychological trauma, grief, and distress among survivors and military personnel.

Directly caused: An anonymous intelligence officer expressed being shocked and traumatized after realizing they authorized a strike on a home where the target was absent, killing innocent families.

Indirectly caused: Widespread trauma, grief, and terror were experienced by Gazan civilians, such as Amjad Al-Yazji, who discovered 11 deceased family members under the rubble of their home.

Inferred additional harm: Widespread, long-term post-traumatic stress disorder (PTSD) and severe psychological trauma are inferred to affect hundreds of thousands of Gazan residents, particularly children.

People affected

• Occurrences reported: 1
• People reportedly harmed: 15000
• People reportedly exposed: 37000

Potential causes

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.

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.

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.

Information quality

• Classification confidence: High
• Reason for confidence: 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.
• Ambiguities identified: The exact mathematical algorithms and specific features used by Lavender to calculate the 1-100 scores remain proprietary and are described only generally. There is also a conflict between the testimonies of the intelligence officers and the official denials of the IDF spokesperson regarding the automated nature of the kill lists.
• Alternative interpretations: The IDF interprets these systems purely as auxiliary databases meant to assist human analysts, rather than automated decision-making systems that direct lethal strikes.

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.

• Overall national security impact: Critical
• Scope: Multiple nations
• Primary target: Palestine
• Other affected: Middle East region
• Alleged perpetrator: Israel

Threat characteristics

• Imminence: Immediate. Active military campaign with ongoing kinetic strikes and immediate life-safety implications.
• Autonomy: Human-supervised. Targeting decisions were generated by AI but required a brief 20-second human verification before execution.
• Novelty: First-of-its-kind. First documented instance of a military utilizing mass-scale AI profiling and tracking systems to direct lethal airstrikes.

Impact by dimension

• Physical security: Critical. AI-directed targeting systems (Lavender and The Gospel) directly authorized kinetic airstrikes resulting in over 15,000 reported deaths and the destruction of 70 percent of Gaza's housing and infrastructure.
• Information security: Severe. Systematic deployment of mass surveillance capabilities harvesting data on 2.3 million residents to feed targeting models, representing an unprecedented scale of automated intelligence collection.
• Sovereignty: Severe. Foreign military forces deployed AI target-generation systems within contested territory, effectively replacing human command decisions with automated targeting algorithms, disrupting sovereign control.
• Economic security: Substantial. Airstrikes on AI-selected targets caused billions of dollars in property and infrastructure damage, decimating the local economy, though strategic global tech security was not directly targeted.
• Societal stability: Critical. Algorithmic profiling of 2.3 million people with a 10 percent error rate led to systemic human rights violations, mass civilian casualties, and the complete collapse of societal stability in Gaza.