IDF's AI targeting system in Gaza raises civilian casualty concerns
Israel's military developed AI systems including "Lavender" and "Gospel" over more than a decade to automate and accelerate targeting in Gaza since October 2023, reportedly processing intelligence at unprecedented speed to bomb 12,000 targets in weeks. Military sources and former personnel expressed concerns that the AI-driven approach significantly increased civilian casualties by reducing verification time and expanding acceptable casualty thresholds.
What happened
Israel developed AI systems including 'Lavender' and 'Gospel' over more than a decade to support military targeting, which became more prominent in 2020 under Unit 8200 leadership.
The AI systems processed intelligence at unprecedented pace, enabling the Israeli military to bomb 12,000 targets in Gaza within weeks after October 7, 2023, with evaluation time reduced to as little as three minutes.
When the pre-compiled target database became depleted, the IDF deployed the 'Habsora' AI system to rapidly generate additional target suggestions, allowing the military to maintain its campaign's pace.
Reported impact
Sources including former Israeli military personnel and intelligence officers expressed concern that the military's use of AI significantly increased civilian casualties.
Internal critics within the IDF argue that AI accelerated the death toll in Gaza, and that Israel's military expanded acceptable civilian casualty thresholds from historic norms, enabled by automation that made it easier to speedily generate large quantities of targets.
Index analysis
A Washington Post investigation revealed debates within the highest echelons of the Israeli military, starting years before October 7, about the quality of intelligence gathered by AI, whether the technologies' recommendations garnered sufficient scrutiny, and if the focus on AI weakened the IDF's intelligence capabilities.
Unit 8200 trained AI to create target databases by cross-referencing billions of signals gathered from drones, fighter jets, and seismic sensors with phone numbers, social media profiles, and other data, though sources stated findings needed verification by two independent sources and constant updates.
Governance controls
At Nevatim Airbase, military commanders restructured operations as 'AI factories' where soldiers worked prioritizing speed over intelligence assessment, with Mossad and Unit 8200 developing a centralized intelligence pool and an app called 'Hunter' giving soldiers direct access to intelligence data.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.