What Israel's use of AI in Gaza revealed about their civilian harm thresholds

A Bloomberg reporter's observation about Israel's deployment of artificial intelligence in Gaza has cracked open an uncomfortable question that the US military has spent decades trying to avoid: how much civilian harm is acceptable in the pursuit of military objectives?

Katrina, a Bloomberg reporter whose full name was not disclosed in the briefing, told GZERO World that Israel's use of AI in the Gaza campaign felt "potentially uncomfortable for the US military tradition." That single sentence captures a tension that runs through years of American doctrine, international law debates, and the technological transformation of warfare. It also raises a direct, pressing question for the Pentagon: if the US were to deploy AI-assisted targeting at the scale Israel has, would American rules of engagement allow it?

What the quote tells us

The phrase "potentially uncomfortable for the US military tradition" is dense with meaning. It suggests that the way Israel used AI was not merely a technical difference — faster targeting, better data fusion — but a departure from something deeply embedded in American military culture. That tradition includes the principle of distinction, the requirement to discriminate between combatants and civilians, and the proportionality rule that prohibits attacks where civilian harm outweighs military gain.

The key word is "threshold." Every military force maintains a civilian harm threshold — sometimes written into law, sometimes embedded in doctrine, sometimes just left to commanders on the ground. The threshold determines how many civilian casualties are considered acceptable in a given strike. The United States has historically treated its threshold as relatively stringent, especially after the wars in Iraq and Afghanistan. Israel's campaign in Gaza, according to the Bloomberg reporter, appears to have operated under different assumptions, and AI made that difference visible.

AI as a magnifying glass

Artificial intelligence did not create the ethical problem of civilian harm thresholds. What AI does is accelerate and scale decisions that used to be made by individual intelligence analysts and targeting officers. A human analyst can assess a limited number of targets per shift. An AI system can process thousands of hours of surveillance footage, cross-reference signals intelligence, and recommend targets in minutes. The technology is not inherently evil, but it can amplify whatever ethical standards are programmed into it — or left out.

The Bloomberg reporter's observation implies that Israel's AI systems were tuned to accept a higher level of civilian risk than the US military tradition would consider acceptable. That could mean AI was used to authorize strikes on targets that a human analyst would have flagged as too dangerous for civilians, or that the speed of AI recommendations overwhelmed the normal vetting process. Without specific details from the source material, we cannot say exactly what Israel did. But the discomfort the reporter described points to a gap between how the two countries think about acceptable harm.

US military tradition: a moving target

It is worth remembering that the "US military tradition" on civilian harm has not been static. The firebombing of Tokyo and the atomic bombs dropped on Hiroshima and Nagasaki during World War II would be indefensible under today's rules of engagement. The Vietnam War included free-fire zones. Even in the post-9/11 period, the US conducted signature strikes in Pakistan and Yemen that killed unknown numbers of civilians. The tradition is not a fixed moral line.

What has changed is public and international scrutiny. From the My Lai massacre to the Abu Ghraib photos to the drone strike accountability debates, the US military has been forced to codify its civilian harm prevention processes. The result is a system of legal reviews, collateral damage estimates, and commander approvals that slow down targeting decisions. AI threatens to bypass that system unless deliberately constrained.

Israel's approach, as described by Katrina, may be less constrained. The Bloomberg reporter suggested the Israeli AI targeting pipeline operated at a speed and scale that felt uncomfortable to American observers. That does not mean US forces would never use AI in a similar way. It means the cultural and doctrinal brakes are different.

What this means for the future of warfare

The Gaza campaign is the first major conflict where AI targeting appears to have been used at scale. The United States and other militaries are watching closely. The Pentagon has its own AI initiatives, including Project Maven and the Joint All-Domain Command and Control system. Defense officials have repeatedly said that humans will remain in the loop for lethal decisions. But the loop can get very short when AI is feeding recommendations faster than any human can review them.

One of the uncomfortable truths the Bloomberg reporter's comment points to is that the US military tradition may not be equipped to handle AI-driven warfare without significant reform. The tradition relies on human judgment, context, and the ability to pause. AI systems are built for speed, pattern recognition, and scale. The two do not naturally coexist.

Civilian harm thresholds are not just legal abstractions. They determine real outcomes for families, neighborhoods, and entire communities. If Israel's use of AI in Gaza lowered its threshold — consciously or through the sheer pace of operations — then the technology itself becomes a tool for expanding the boundaries of acceptable destruction. That is the "potentially uncomfortable" reality for a US military that has spent years drawing lines to prevent that expansion.

A question of accountability

Another dimension of the Bloomberg reporter's remark concerns accountability. In the US system, individual commanders are held responsible for targeting decisions. If an AI system recommends a strike that kills civilians, the legal and ethical burden falls on the person who authorized it. But if the AI system was trained on data that already normalizes high civilian casualties, or if it operates at a speed that makes meaningful human review impossible, the accountability chain breaks.

Israel has not publicly detailed exactly how its AI targeting systems function, nor has it disclosed the civilian casualty numbers from the Gaza campaign. Independent estimates vary widely. The Bloomberg reporter's observation does not include specific figures. What it does is signal a concern shared by observers inside and outside the US defense establishment: that the use of AI in Gaza may represent a new normal for civilian harm, one that the US military tradition was not designed to accommodate.

Conclusion

Katrina's comment to GZERO World is a single data point, but it comes from a credentialed Bloomberg reporter and it addresses a subject of extraordinary consequence. Israel's use of AI in Gaza has become a case study in how fast, automated targeting can change the moral and operational calculus of war. The "potentially uncomfortable" nature of that use for the US military tradition is not an indictment of Israel. It is a warning for the United States.

The Pentagon cannot un-learn what Israel's campaign has demonstrated. AI targeting works – it finds targets faster, processes more intelligence, and generates more options. The question is whether the US military tradition of civilian harm prevention can adapt to that speed without breaking. The Bloomberg reporter's quote suggests that it cannot, not without rethinking what the threshold really is.

That conversation needs to happen openly, before US AI systems are fielded in the next conflict. The technology is here. The doctrine has not caught up. And the discomfort the reporter described should be a starting point for that debate, not an afterthought.