IDF Collateral Damage AI System Sparks Debate Over Gaza Targeting
Documentary 'NAZA' examines Israeli military's use of artificial intelligence to calculate civilian harm, raising questions about thresholds and accuracy rather than the practice itself.
The controversy behind NAZA
A documentary premiering at the Venice Film Festival has reignited debate over how Israel's military uses artificial intelligence in Gaza operations—but the actual dispute may be narrower than headlines suggest.
"NAZA," directed by Israeli filmmakers Yuval Abraham and Rachel Szor, takes its title from the Hebrew military abbreviation for collateral damage: nezek agavi. The 80-minute film, which won Venice's Special Jury Prize in September 2026, examines AI systems the Israel Defense Forces uses to assess expected civilian harm before strikes.
The central tension is not whether Israel calculates collateral damage—both the IDF and the filmmakers' own reporting confirm it does—but what those calculations mean and whether the limits placed on civilian harm were adequate.
How collateral damage assessment works
Cyber intelligence expert Matías Rott explained to The Media Line that collateral damage refers to unintended harm beyond a military target. AI systems help commanders model a target's surroundings, assess blast effects from different attack angles, and evaluate weapon choices to minimize incidental casualties.
"When you are working with this AI, for sure what you are willing or intending is to minimize collateral damages," Rott said. The technology can process large intelligence volumes and identify, for example, whether striking a building from one direction would affect nearby structures while another approach faces open space.
According to published IDF procedures, strikes should not proceed if expected collateral damage is excessive relative to the anticipated military advantage. The assessment exists as a constraint, not a license.
The real allegations
The documentary's more serious claims concern thresholds and accuracy. According to Abraham's earlier reporting for +972 Magazine in April 2024, intelligence sources alleged that during early weeks of the Gaza war, attacks against some suspected low-ranking operatives could be authorized with an anticipated civilian toll of up to 15 or 20 people. The IDF disputed this, saying it assesses targets individually.
Sources also questioned whether automated estimates of civilian presence in buildings relied on prewar residency data adjusted by broad evacuation assumptions, rather than real-time confirmation of who remained inside.
The film examines systems including Lavender, which processes intelligence on suspected Hamas and Palestinian Islamic Jihad operatives, and Habsora, which flags potentially relevant objects for analysts. The IDF describes these as decision-support tools that cannot independently designate targets—human commanders must still authorize strikes.
Rott identified a legitimate concern: whether human operators genuinely evaluate AI outputs or function as "rubber stamps" for automated recommendations.
Why it matters
The debate over military AI often conflates the existence of algorithmic targeting assistance with autonomous weapons. This case illustrates a more nuanced reality: AI systems that calculate collateral damage can be designed to reduce civilian harm while still facing valid criticism about threshold-setting, data accuracy, and human oversight quality. For defense organizations and technology companies developing similar capabilities, the distinction matters—the controversy is not that harm is measured, but how those measurements inform decisions and whether the constraints are sufficient.
IDF response
The Israeli military has categorically rejected the documentary's allegations and challenged reliance on anonymous sources whose positions and access cannot be verified. IDF Chief of Staff Lt. Gen. Eyal Zamir ordered an examination of whether classified material was leaked.
The filmmakers' case will be easier to assess when "NAZA" receives wider release. For now, the terminology itself clarifies the dispute: militaries need collateral damage estimates not to know how effectively they can destroy targets, but to decide what else may be destroyed and whether strikes should proceed.
These details were first reported by Gabriel Colodro for The Media Line and published in The Jerusalem Post.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
