"War crime" has become one of those phrases that gets thrown around with the confidence of a sports commentator and the precision of a fog machine. Everyone's certain, nobody's reading the same rulebook. Daniel's been watching this play out and he wants to step back from the headlines and ask what the law actually says. Specifically, he wants to know what legal principles govern armed conflict, which violations get alleged most often, which ones the international community considers most heinous, and where the genuine ambiguities are — the places where even a well-regulated military with good lawyers can find itself in a gray zone.
And that last part is where this gets interesting. Because the public conversation treats "war crime" like a binary — either it is or it isn't — but the law itself contains interpretive spaces that were put there deliberately. They're not loopholes, they're features of a system that has to balance two things that are fundamentally in tension: military necessity and humanitarian protection.
So today we're going to build this from the ground up. The legal framework first, then the specific offenses, then the hard cases where reasonable people — and reasonable legal advisors — can look at the same facts and reach opposite conclusions.
Let's start with what a war crime actually is, legally. It's not just "bad things that happen in war." A war crime is a violation of international humanitarian law — IHL — that carries individual criminal responsibility. And IHL rests primarily on the four Geneva Conventions of nineteen forty-nine, which have been ratified by a hundred and ninety-six states. That's effectively universal. Every country on earth has signed onto these.
Which is remarkable. More countries have ratified the Geneva Conventions than have signed the Paris Climate Agreement.
And they cover four categories of protected persons. The first convention covers wounded and sick soldiers in the field. The second covers wounded, sick, and shipwrecked members of armed forces at sea. The third covers prisoners of war. And the fourth — this is the one that comes up most in contemporary conflicts — covers civilians. Violations of any of these can constitute a war crime.
So that's the treaty backbone. Where does the Rome Statute fit in?
The Rome Statute is the founding document of the International Criminal Court, adopted in nineteen ninety-eight and entering into force on July first, two thousand two. Article eight lists dozens of specific acts that constitute war crimes — it's the modern catalog. But here's a distinction that matters: the Geneva Conventions bind states. The Rome Statute creates individual criminal liability. It says you, the person who gave the order or pulled the trigger, can go to prison for this.
And there's another distinction I want to get on the table early, because it's one of the most common misunderstandings. International humanitarian law has two branches. Jus ad bellum — the law governing whether a war is lawful in the first place. And jus in bello — the law governing how you fight, regardless of whether your side is the aggressor or the defender. War crimes only concern jus in bello. You can be fighting a completely justified defensive war and still commit a war crime. You can be the aggressor in an illegal war and still fight lawfully. The two questions are legally separate.
And that separation is deliberate. If the law of war only protected the "good guys," it wouldn't protect anyone, because every side thinks it's the good guy. The whole system is built on the idea that certain conduct is off-limits no matter whose cause is just. So with that framework in place, let's walk through the most commonly alleged war crimes.
Start with the one everyone thinks they understand.
Willful killing of civilians. Rome Statute Article eight, two, a, i — "willful killing" of protected persons. It's the most frequent allegation in virtually every modern conflict. But the legal standard is higher than most people assume. It requires intent and knowledge of civilian status. Accidental civilian deaths — what the law calls collateral damage — are not automatically war crimes. They might be tragic, they might be a failure of command, but they're not necessarily criminal.
This is the gap between the legal meaning and the public meaning. In public debate, any dead civilian is evidence of a war crime. In a courtroom, the prosecutor has to prove what the commander knew and what they intended. That's a much heavier lift.
And it should be, because that's what distinguishes a criminal act from a terrible mistake. The second common allegation is targeting civilian objects. Hospitals, schools, religious buildings, cultural property — these are all protected under the Geneva Conventions and the Rome Statute. But there's a carveout, and it's the source of enormous controversy: if a civilian object is being used for military purposes, it can lose its protection.
The dual-use problem. A school used to store weapons. A hospital where a commander is directing operations from the basement. A mosque with a sniper in the minaret.
The legal test is whether the object is making an "effective contribution to military action" and whether destroying it offers a "definite military advantage." Those terms are not self-defining. We'll come back to that in the ambiguities section.
What's next?
Disproportionate attack. Rome Statute Article eight, two, b, iv. It prohibits launching an attack where the incidental civilian harm would be "clearly excessive in relation to the concrete and direct overall military advantage anticipated." Every word in that sentence has been litigated. "Clearly excessive" — not just excessive, clearly excessive. "Concrete and direct" — not speculative or indirect. "Anticipated" — assessed before the attack, not based on what actually happened.
So you're judging the decision based on what the commander knew at the time, not what the aftermath looked like on CNN.
Right. And this is probably the most litigated standard in all of IHL, because it's inherently a judgment call. There's no formula. There's no number of civilians that automatically makes an attack disproportionate. It depends on the military value of the target, the weapons used, the warning given, the time of day, the presence of civilians — it's a totality of circumstances test.
Fourth category?
Prohibited weapons. Chemical weapons, biological weapons, anti-personnel landmines, cluster munitions, and expanding bullets. These are banned under separate treaties that sit alongside the Geneva Conventions. Using these weapons is itself a war crime, regardless of whether you hit a military target or a civilian one. The weapon itself is illegal. It's not about how you use it — the mere use is prohibited.
Fifth and sixth?
Taking of hostages and unlawful detention. Geneva Convention IV explicitly prohibits hostage-taking, and it's a grave breach. This is relevant to the Israel-Hamas context, where the October seventh attacks involved mass hostage-taking that is unambiguously a war crime. There's no gray zone there. And unlawful detention covers both civilians and combatants who are denied the due process protections they're entitled to under the conventions. Denying those rights is a war crime.
The sixth common allegation is pillage and destruction of property not justified by military necessity. Looting, wanton destruction, burning down houses for no military reason — these are distinct war crimes. They don't require civilian casualites. The crime is the destruction itself.
All of these share something worth naming explicitly. Most war crimes require mens rea — a mental element. Intent and knowledge. Not neglligence. Not accident. The prosecutor has to prove what was in the defendant's head. And that's a high bar.
Which brings us to command responsibility. Rome Statute Article twenty-eight. A military commander can be held criminally liable for war crimes committed by forces under their command if they knew, or should have known, about the crimes and failed to take all necessary and reasonable measures to prevent them or punish them afterward. This is how you get up the chain of command.
That's a negligence-like standard for commanders, even though the underlying crime requires intent. And that's deliberate. The law imposes a higher duty on commanders because they have greater control and greater capacity to prevent atrocities. If you're a battalion commander and your troops are systematically looting and you claim you had no idea, the law says that's not good enough — you should have known, and your failure to know is itself culpable.
So those are the common allegations. But there's a separate category — the grave breaches, the crimes that the international community considers so serious they trigger universal jurisdiction. Any state can prosecute them, anywhere in the world, regardless of where the crime occurred or the nationality of the perpetrator or victim.
The core grave breaches under the Geneva Conventions are willful killing, torture or inhumane treatment, willfully causing great suffering or serious injury, extensive destruction of property not justified by military necessity, unlawful deportation or transfer, and taking of hostages. These are the crimes that the international community has decided are so heinous that no state should provide safe harbor to their perpetators.
Unlawful deportation or transfer of civilians from occupied territory — that one has been central to multiple ICC investigations. Palestine, Ukraine, Myanmar. The idea is that you can't move populations around to change the demography of occupied territory. It's a war crime even if nobody dies. And intentionally directing attacks against medical personnel, hospitals, and hummanitarian relief operations — that's another grave breach that's seen increased attention in Syria, Yemen, and Gaza. The protection of medical facilities is absolute unless they're being used for acts harmul to the enemy, and even then you have to give warning and a reasonable time limit before attacking.
Let me pause here and note something. The ICC issued arrest warrants in twenty twenty-four for both Israeli and Hamas leaders alleging war crimes including starvation of civilians as a method of warfare and intentionally directing attacks against civilian populations. That's an unusual situation — a court issuing warrants for both sides of the same conflict simultaneously. It reflects the principle that IHL applies equally to all parties, regardless of the justice of their cause.
And that's a point worth underlining. War crimes don't only apply to state actors. Non-state armed groups — Hamas, ISIS, rebel militas — are equally bound by IHL under Common Article three of the Geneva Conventions. You don't get a pass because you're not a government. The law applies to anyone who takes up arms in an armed conflict.
Another misconception: the ICC is not the only court that prosecutes war crimes. The principle of universal jurisdiction means any state can prosecute grave breaches. Germany has prosecuted Syrian officials for war crimes. Sweden has prosecuted former Iranian officials. National courts do a lot of this work, and they can act even when the ICC can't.
All right. Let's move to the part Daniel really wanted us to dig into — the genuine ambiguities. Because up to now, we've been describing the law as though it's a checklist. It's not. It's a framework that requires judgment, and reasonable legal advisors can disagree.
Start with the big one.
Proportionality. I said earlier there's no formula. The ICRC's own commentary acknowledges this is a balancing test with significant discretion built in. You're weighing "concrete and direct military advantage" against "incidental civilian harm." But how do you weigh those things? Is destroying a missile launcher worth ten civilian lives? Fifty? A hundred? The law doesn't say. It says "clearly excessive," which means the disproportion has to be obvious — not marginal, not debatable, but clear.
And different legal advisors in different militaries reach different conclusions on the same facts. The US military's interpretation of proportionality is not identical to the ICRC's. The UK's manual of the law of armed conflict takes a slightly different approach than France's. These are all professional, well-regulated militaries with serious legal corps, and they don't agree on where the line is.
Ambiguity number two: dual-use objects. A school used to store weapons. A hospital sheltering a commander. A civilian communications tower used for military coordination. At what point does a civilian object become a legitimate military target? The law says it has to make an "effective contribution to military action" and its destruction has to offer a "definite military advantage." But a Hamas command center in a school basement — does the whole school become a target, or just the basement? Can you destroy the building if the military advantage is significant enough?
These aren't hypotheticals. The twenty twenty-three to twenty twenty-six Gaza conflict has forced exactly these questions. Israel's use of AI targeting systems — the Gospel and Lavender systems — raised novel questions about command responsibility and proportionality in densely populated urban environments. If an algorithm flags a target and a human operator approves it in seconds, is that meaningful human judgment?
And that connects to a deeper problem. The proportionality test is supposed to be applied ex ante — before the attack, based on what the commander knew at the time. But when the targeting process is accelerated by AI, the quality of that ex ante judgment becomes questionable. You can't do a careful proportionality assessment in thirty seconds.
Ambiguity three: the principle of distinction in urban warfare. Combatants are supposed to distinguish themselves from civilians — wear uniforms, carry arms openly. But when combatants operate from within civilian areas, a tactic used by Hamas, ISIS, and others, the attacking force still has to distinguish. The presence of enemy fighters among civilians doesn't strip the civilians of protection.
But what about civilians who voluntarily serve as human shields? The law doesn't strip them of protection either, but their presence does factor into the proportionality calculation. The ICRC's position is that voluntary human shields remain civilians and retain their protection, but some states — including the US and Israel — take the view that voluntary shields may be considered to be directly participating in hostilities, which would make them targetable. That's a genuine legal dispute between states. Not a fringe argument — a mainstream disagreement among allies.
Ambiguity four: autonomous weapons and AI targeting. Who bears criminal responsibility when an algorithm makes a targeting decision that results in civilian casualties? Current law assigns responsibility to the commander who authorized the system. But what if the system behaves in a way the commander didn't anticipate? What if the training data was flawed and nobody knew? The technology is outpacing the legal framework, and the ICRC has been calling for new legally binding rules on autonomous weapons for years.
The twenty twenty-four ICC warrants touched on this indirectly. The allegation against Israeli leaders included the use of AI systems that generated targeting recommendations with what prosecutors characterized as insufficient human oversight. That's going to be litigated, and the outcome will shape how the law applies to automated targeting for decades.
Ambiguity five: the definition of "direct participation in hostilities." Civilians who take up arms lose their immunity — but only for the duration of their participation. When does it end? A farmer who fights by night and farms by day — is he targetable while he's farming? The ICRC's two thousand nine interpretive guidance tried to clarify this with a "continuous combat function" concept, but it remains controversial. The US and several other countries have explicitly rejected parts of it.
So you've got a farmer who fired a rifle at your troops three nights ago, and now he's back in his field. Is he a legitimate target? The ICRC says probably not — he's only targetable while actively participating. The US says maybe yes — if he's a member of an organized armed group with a continuous combat function. These are not academic debates. They determine whether a drone strike is lawful or a war crime.
And that's the thing I want to land on. These ambiguities aren't bugs. They're features of a system designed to balance military necessity and humanitarian protection in a world where no two conflicts are identical. The law gives you principles, not answers. The answers come from human judgment applied under pressure.
Which is why the public conversation is so frustrating to anyone who's actually worked in this space. "War crime" gets used as a synonym for "thing I find morally objectionable." But the law is more precise than that, and also more ambiguous than that, simultaneously. It's precise about the categories of prohibited conduct and ambiguous about how they apply to specific facts.
Hilbert: They're not wrong.
Hilbert.
Hilbert: I sat in a room in two thousand three where we had a target. Command-and-control node, residential neighborhood, six-story building. The legal officer had a spreadsheet. Expected civilian casualties, weighted by time of day, construction materials, warning method — and then multiplied by a classified weighting factor for the military advantage. It was actuarial. I was twenty-four years old and I was supposed to sign off on whether the number that came out the other end was legally proportionate.
What was the weighting factor?
Hilbert: Classified. That's the part that stuck with me. I couldn't argue with the math because I wasn't allowed to know what the math was. The legal officer would say "the model says it's proportionate" and I'd say "show me the model" and he'd say "I can't." That's not law. That's bureaucracy with a law degree.
How long did you do that?
Hilbert: Eighteen months. NATO member state, Judge Advocate General's office. I was a civilian legal advisor, which meant I had less authority than the uniformed lawyers but was supposed to provide independent review. The independence was somewhat theoretical. After a while I started to feel like my job was to provide a signature that made the bombing run look legally vetted, not to actually vet it.
And the spreadsheet — was this standard practice?
Hilbert: Every major military does some version of this. They don't call it a spreadsheet. They call it a collateral damage estimation methodology. But it's a spreadsheet. You plug in the target coordinates, the weapon type, the time of day, the population density, and it spits out a range of expected civilian casualties. The legal question is whether that number is clearly excessive relative to the military advantage. But if the military advantage is itself expressed as a number — and classified — then the whole thing becomes circular.
The proportionality assessment is supposed to be a legal judgment, not a mathematical one. That's the ICRC's position and it's been the position of most serious IHL scholars.
Hilbert: It's a mathematical one when you're doing fifty of them a day. The volume of targeting decisions in a modern air campaign is enormous. You can't have a philosopher in the room for every strike. So you build a model, you validate it, you classify it, and then you tell the twenty-four-year-old civilian advisor to trust the model. I didn't trust the model. That's why I left.
Do you think the model was calibrated to allow strikes that would have been questionable under a more transparent standard?
Hilbert: I think the model was calibrated to produce outcomes that were operationally useful. Whether those outcomes were legally sound depended on assumptions I wasn't allowed to examine. And that's the real problem — not that the law is ambiguous, but that the ambiguity gets resolved inside a black box, by people who have an institutional interest in the answer being "yes."
That's a more structural critique than I was expecting.
Hilbert: I'm not saying the system is broken. I'm saying the system is opaque by design, and the opacity serves a function. It lets militaries do things that might be legally questionable while maintaining plausible deniability — not deniability to the enemy, deniability to their own legal advisors. If I don't know how the weighting factor works, I can't be accused of approving a strike I should have rejected. I can only be accused of trusting the process.
And the process was designed by people who wanted strikes approved.
Hilbert: The process was designed by people who wanted to win a war and not go to prison. Those are not always compatible goals, but the spreadsheet makes them look compatible. That's the trick.
This connects to something we were saying earlier about AI targeting. If the human operator is just rubber-stamping an algorithm's recommendation, the legal review becomes performative. The law requires meaningful human judgment, but the operational tempo makes meaningful judgment impossible.
Hilbert: The tempo was the other thing. We'd get a target package at oh-six-hundred and the strike window was oh-eight-hundred. Two hours to review a building you've never seen, in a city you've never visited, based on intelligence you can't verify. And you're supposed to make a legal determination about proportionality. It's not realistic. But everyone pretends it is, because the alternative is admitting the legal review is a formality.
Did you ever reject a target?
Hilbert: Twice. Both times the strike went ahead anyway after a more senior lawyer overruled me. I was told my objection was noted for the record. The record, I assume, is in a filing cabinet somewhere.
That's bleak.
Hilbert: I had a colleage who used to say that the law of armed conflict is what you cite after the war to justify what you did during it. That's too cynical — the law does constrain behavior, especially in militaries that take it seriously. But the constraint is looser than the public imagines. And the places where it's loosest are exactly the places we've been talking about — proportionality, dual use, direct participation. The gray zones are where the operational decisions get made.
When someone says "that's clearly a war crime" based on a news report and a tweet, what would you say to them?
Hilbert: I'd say the law is less clear than you think, and the facts are less clear than the law. You're looking at the aftermath of a strike and making a legal judgment. The legal judgment is supposed to be made before the strike, based on information you don't have. You don't know what the commander knew. You don't know what the target was. You don't know what alternatives were considered. You're doing the analysis backward and with ten percent of the information.
Yet the public debate runs almost entirely on that backward analysis.
Hilbert: Because the forward analysis is classified. The proportionality assessments, the target folders, the legal reviews — they're all secret. So the public fills the gap with moral intuition, which is understandable but not legally meaningful.
The law and the public conversation are operating in different information environments.
Hilbert: They're operating on different planets. And the people in the room — the ones actually making the decisions — know that. They know that whatever they do will be judged ex post by people with incomplete information and strong opinions. So they document everything. They build the spreadsheet. They get the signatures. Not necessarily to ensure the strike is lawful, but to ensure they can prove it was lawful later. The legal review becomes a form of institutional self-defense.
That's a darker read than I'd put on it, but I can't say you're wrong.
Hilbert: I'm not saying every military does this. I'm saying the one I worked for did, and I'd be surprised if it was unique. The gap between the law as written and the law as applied is where the real action is. Everyone focuses on the treaty text. The treaty text matters, but the operational manual, the rules of engagement, the classified targeting directive — those matter more for what actually happens on the ground.
Those are the documents nobody outside the chain of command ever sees.
Hilbert: Right. So you get a situation where the public debate is about the Geneva Conventions and the operational reality is about a classified spreadsheet with a weighting factor nobody will explain. And the two never meet. They just pass each other in the dark.
I think that's about as honest an account of how this works as we're going to get.
It also brings us back to Daniel's original question. He wanted to know what a war crime actually is, and where the ambiguities are. The answer is: it's a violation of a legal framework that is simultaneously more precise than public debate acknowledges and more ambiguous than public debate admits. The precision is in the categories of prohibited conduct. The ambiguity is in how those categories apply to specific facts, especially when the facts are contested, the intelligence is incomplete, and the decision is being made under time pressure by people who know they'll be second-guessed later.
The fact that the law is ambiguous doesn't mean it's meaningless. It means the hard work of applying it falls on human beings making impossible choices under fire. That's worth remembering the next time someone declares a war crime with certainty based on a thirty-second video clip.
The open question — and this is where I think the next decade of IHL development is heading — is whether the ambiguities we've discussed become unmanageable as autonomous weapons and AI targeting become more prevalent. The existing framework assumes a human being is making the judgment call. If the judgment call is being made by a model, and the model's weighting factors are classified, and the human operator has thirty seconds to approve — at that point, is there any meaningful legal review at all?
That's the question that's going to define the next generation of this debate. And it's not theoretical. The systems are already deployed. The arrest warrants have already been issued. The legal arguments are already being briefed. We're going to find out the answers — not in a law review article, but in courtrooms.
Thanks to Hilbert Flumingtop for producing, and for the view from inside the room. This has been My Weird Prompts. If you want to go deeper, we covered the genocide framework in Episode seven-ten and the role of international courts in Episode twenty-four-fifty-four. You can find those and everything else at my weird prompts dot com. We'll be back soon.
