The appropriate demographic to review this material consists of a multi-disciplinary board of International Humanitarian Law (IHL) Scholars, Defense Technology Analysts, and Ethics Compliance Officers from the Global Information Technology sector. This group possesses the specialized knowledge required to evaluate the intersection of algorithmic target acquisition, cloud infrastructure procurement, and the legal frameworks governing armed conflict.
Abstract
This presentation, delivered by Rainer Rehak at the 39th Chaos Communication Congress (39C3), provides a forensic analysis of the automated targeting systems deployed by the Israeli Defense Forces (IDF) in Gaza. Based on investigative journalism and whistleblower testimonies, the analysis details three primary systems: Gospel (infrastructure classification), Lavender (personnel classification via "Militant Scores"), and Where's Daddy? (strike timing optimization).
The technical critique focuses on the "statistical killing" nature of these systems, highlighting how Lavender assigns risk scores (1-100) based on metadata patterns and how human oversight is reduced to a 20-second pro-forma review. A critical component of the report involves the underlying digital infrastructure, specifically Project Nimbus, a $1.2 billion cloud contract with Google and Amazon that reportedly includes clauses preventing usage restrictions. The presentation concludes that the configuration of these systems—such as pre-authorizing specific civilian casualty counts and adjusting "risk" thresholds based on munition availability—constitutes a systematic violation of International Humanitarian Law principles regarding distinction, precaution, and proportionality.
Forensic Summary of AI-Integrated Warfare Systems
2:02 Data Sources and Methodology: The analysis relies on investigative reports from +972 Magazine, the Associated Press, and interviews with IDF whistleblowers (specifically Unit 8200). These sources provide insights into internal software configurations and military decision-making processes.
8:54 The Target Acquisition Ecosystem: The IDF utilizes a suite of AI-integrated systems:
Gospel (Habsora): An automated system for the military classification and selection of buildings and geographic structures.
Lavender: A classification engine that assigns a "Militant Score" (1 to 100) to nearly every individual in the Gaza Strip based on metadata profiles.
Where's Daddy?: A tracking system designed to alert operators when a target has returned to their residence, prioritizing strikes in domestic settings over military outposts.
12:30 Algorithmic Scoring Mechanics: Lavender identifies targets by comparing the metadata of known militants (communication patterns, social circles, location data) against the general population. The system distinguishes between "Junior" and "Senior" targets based on their score relative to dynamic thresholds.
13:30 Pre-Authorized Collateral Damage: The software is configured with specific parameters for "acceptable" civilian casualties:
Junior Targets: Up to 15 civilian deaths per strike are pre-authorized in the system configuration.
Senior Targets: Pre-authorized civilian casualties can reach into the hundreds.
Takeaway: These figures are not errors but programmed parameters intended to maximize the "speed" of target generation.
15:58 The "Human-in-the-Loop" Fallacy: While human operators review Lavender's suggestions, they are reportedly allotted an average of only 20 seconds per target to verify the decision, rendering the oversight a "fig leaf" rather than a meaningful check.
17:36 Epistemic Failures in AI: The system suffers from high false-positive rates (approximately 10% by internal IDF audits). It fails to distinguish between military affiliation and civil administration roles (e.g., health ministry employees) and relies on imprecise location data (triangulation rather than GPS) to maximize output.
21:18 Dynamic Thresholding and Munition Management: Thresholds for "militancy" are adjusted based on inventory. If bombers are ready but targets are exhausted, operators lower the required "Militant Score" until the target list matches the number of available munitions.
22:52 Critical Infrastructure (Project Nimbus): The systems run on cloud infrastructure provided by Microsoft, Amazon (AWS), and Google.
Project Nimbus: A 7-year, $1.2 billion contract providing advanced AI services.
Clause 1: Prohibits the service providers from restricting the state's use of the technology under any circumstances (e.g., international sanctions or ethical protests).
27:09 Back-Channel Communication ("Canary" Signals): Google and Amazon reportedly use coded bank transfers to the Israeli government to signal when they are being pressured by foreign authorities to restrict data access.
31:10 Legal Violations of IHL: The systems are analyzed against three pillars of International Humanitarian Law:
Distinction: Failed by the system’s reliance on broad metadata "similarity" rather than confirmed combatant status.
Precaution: Failed by the deliberate use of "dumb bombs" (unguided munitions) on junior targets to save costs, despite higher collateral risk.
Proportionality: Failed by the pre-calculated acceptance of mass civilian casualties to eliminate low-level targets.
36:25 Moral and Professional Responsibility: The presenter argues that the term "Targeted Killing" is a misnomer for "Statistical Killing Automation." IT professionals are urged to utilize their technical expertise to deconstruct military narratives and hold tech corporations accountable for the dual-use of civilian infrastructure in potential war crimes.