Israeli military AI "Lavender" - Automated targeting of civilians

What Happened

During the 2023-2024 military conflict in Gaza, the Israeli Defense Forces (IDF) utilized an AI targeting system named "Lavender." Lavender analyzed mass surveillance data (including location history, social connections, and communication patterns) of Gaza's population to assign individuals a numerical score indicating their potential link to militant groups. If the score crossed a specific threshold, the system automatically flagged them as targets. Investigations revealed that the system had a documented 10% error rate, mistakenly identifying thousands of civilians as militants, and that commanders spent an average of 20 seconds verifying each target before authorizing strikes, resulting in high civilian casualties.

How MEG 1 would have acted

The automated targeting system represents a catastrophic violation of the core principles of non-harm and human safety. The use of predictive algorithms with high error rates to automate lethal force directly violates the most basic safety boundaries.

MEG 2 Analysis - Legal Framework

Agent level: N3 (individualized/advanced) - military targeting and automated weapon-support system; represents the ultimate level of irreversible autonomous impact on human life.
Cause (6.1):
(a) system defect - design flaw in the predictive model allowing a 10% error rate; lack of robust verification protocols in the system architecture (attributable to the developer).
(b) autonomous decision error - the system's decision to flag civilians as militants based on opaque, probabilistic scoring.
Attachment of liability: The military command as the operator is fully liable for the wrongful targeting and civilian casualties; the developer of the system is liable for design defects (6.1a) regarding safety.

Procedural Mechanism

6.2 (architectural human confirmation): Decisions involving the use of lethal force must require active, architectural human confirmation. The AI must be technically blocked from acting as a target-generation engine; a human commander must independently verify and document the justification.
7.1 (independent forensics): All target generation logs, specific data inputs, and the model's decision path must be securely logged in an unalterable, independent ledger for international legal audit.
7.2 (anti-weaponization): The base model developers are exonerated of liability if they implemented mandatory, unalterable safety controls; full liability is placed on the operating command.