Human shields and civilian deaths

The effects of weakened IHL enforcement on the use and abuse of human shields

Macartan Humphreys (WZB)

2026-10-08

Motivation

Launching rockets or hiding weapons near civilians is reprehensible. You traumatize people by using weapons next to them. If you know that the other side is likely to return fire, you either intend that people be harmed for a public relations gain, or you’re at least indifferent to it. If you think that the other side won’t respond because of your proximity to civilians, you are using those people as human shields. […] But to argue that this absolves the other party from responsibility is to get both law and morality wrong

Lieblich (2021)

This project

A (formal) theoretical investigation into the strategic logic of the use of human shields, and the conditions that explain them

Particular interest is the role of sanctions, given the strategic considerations:

  • IHL forbids the use of civilians as shields and also the disregard of shields: the (disproportionate) killing of protected persons that are employed as shields (Geneva Convention IV, Article 28; Additional Protocol I, Article 51(7)–(8) (Sandoz et al. 1987))

  • The question is what those two prohibitions do to who mingles and who dies.

Camouflage

I was stationed […] with another twenty-five soldiers and about twenty Kamajors. […] We heard a lot of noise; people shouting, civilians, children, the sound of people walking and saw what I estimated to be more than 2,000 civilians carrying things on their heads and backs. Suddenly we saw an armed man in the crowd and said, who goes there. He refused to say anything and someone fired a shot and then the whole thing started. […] In the end I think many civilians were killed. […] And the civilians kept coming. They were so many and we just didn’t know who was among them.

ECOMOG soldier in Sierra Leone, quoted by Dufka (1999)

Deterrence

Jesús Abad Colorado, Colegio INEM, Medellín, 1999

Jesús Abad Colorado, 1999. Colegio INEM, Medellín, Antioquia

Colombia: Sayed, Leiden Journal of International Law

Deterrence

Civilians held in metal cages

HRW: The armed groups say they are placing civilians in metal cages to deter indiscriminate government attacks on the area.

Syria; Human Rights Watch, 2 November 2015 (Human Rights Watch 2015)

Subtler forms

Proximate shields, coerced shields, volunteers (Gordon and Perugini 2020).

BBC On This Day, 23 August 1990 Hussein on Irawi television after Kuwait invasions

Jenin hospital raid, 2024

Jenin hospital raid, 2024 (CNN)

Exposure and baiting

Alan J. Kuperman, Darfur: Strategic Victimhood Strikes Again?, Genocide Studies and Prevention, 2009

Rebels place their own civilians at risk in order to draw outside support.

Alan J. Kuperman, Genocide Studies and Prevention 4(3), 2009 (Kuperman 2009)

Varying autonomy

Solomko, Human shield Sergey Solomko, Human shield

Varying autonomy

Solomko, Human shield Sergey Solomko, Human shield

Ernie O'Malley and Sheila Humphreys

Ernie O’Malley and Sheila Humphreys

Opposite timing

New York Times, 14 October 2024: How Israel's Army Uses Palestinians as Human Shields in Gaza

In cases in which shields are used against booby traps, the timing is opposite: the employer of the shield is the second mover.

Natan Odenheimer, Bilal Shbair, and Patrick Kingsley, The New York Times, 14 October 2024 (Odenheimer et al. 2024)

What counts

Ministry of Foreign Affairs, 11 July 2014: excerpt treating a call to gather on a roof as the use of human shields

What counts is disputed. Civic resistance or human shields?

Ministry of Foreign Affairs, Hamas again uses Gazan civilians as human shields, 11 July 2014 (Ministry of Foreign Affairs 2014)

Varying scales

BBC

BBC: Gaza October 2025

“What could change the equation was a great act, and without a doubt, it was known that the reaction to this great act would be big,” Mr. al-Hayya said. NYT (Hubbard and Abi-Habib 2023)

(Though note that this statement does not distinguish intent from byproduct.)

Normative clarifications

I am not dealing directly with the discursive debates on the term, important as they are.

The discursive construction of civilians as human shields supposedly allows a slackening of the rules of proportionality and distinction and legitimises a higher level of collateral damage by transferring responsibility for civilian losses to enemy forces and sometimes even to the people themselves if they are presumed to have consented to be used as shields. Douillard (2021)

  • The term human shields is sometimes used to suggest that civilians are not protected (Sayed 2024).

    • Law seems quite explicit: Using civilians as shields does not remove their protection.
    • The consequences of this debate can be handled in this framework.

Other normative dimensions

  • The model presupposes that every side puts a nonnegative value on civilian life.

    • That is not necessarily true.

  • The model presupposes that every side benefits from audiences believing that it cares about civilian life.

    • That is not necessarily true. Being seen to care can be treated as a weakness.

A simple game

  • We imagine a world in which a rebel group, R decides whether to mingle among civilians (or remain in ‘the bush’)
  • A state, S, decides whether to attack despite the predictable civilian casualties

Bare bones logic

Simplest model focused on zero-sum components and penalties; abstracting from reputational, value based, and military considerations.

Player 1 (rebels) chooses whether to enter.

  • If they stay out, both get 0.

  • If they enter, Player 2 (the state) must transfer $1 one way or the other.

Player 2 (state) chooses:

  • Fire: take $1 from 1, but pay \pi_2 back to 1.
  • Hold fire: give $1 to 1.

IHL:

  • \pi_1: the sanction on using a shield.
  • \pi_2 is the sanction on disregarding shields.

The game tree

The state will fire iff \pi_2 < 2 (prefers 1-\pi_2 to -1).

Given this, rebels will enter iff that reply beats 0: iff \pi_2 > 1+\pi_1 when the state fires, and iff 1-\pi_1>0 when it holds fire.

Analysis:

Path Later name \pi_2
Stay out Mingling deterred Low. Entering would be met by fire, so rebels stay out.
Enter, then fire Asymmetric exposure (entrapment) Intermediate. The state still fires; rebels enter and collect \pi_2.
Enter, then hold fire Hiding (brinkmanship) High. The state will not fire. Rebels enter.

Raising the sanction on fire can move the path from nobody enters to enter and get shot.

Sanction effects are ambiguous

In this game, a higher \pi_2 can raise the chance of expropriation / violence.

The remedy on this tree is \pi_1. A sanction on entry of \pi_1 \ge 1 shuts the entrapment outcome.

Lessons from tihs example carry into the full model:

  1. A sanction on the second mover is a transfer to the first mover.
  2. That transfer can create the very violence it is meant to deter.
  3. The worst outcomes for civilians can be when \pi_L is low but \pi_H is intermdiate

Full model

What gets added

Varying motives: Each side is a low type or a high type. Low types do not pay for a civilian death. High types do.

The two sides know the pair of types. Civilians do not. Civilians update from the actions they see.

Setting

  • k: rebel identifiability. Higher k is a more discriminating strike.
  • q: death risk in the bush.
  • \pi_S: state-side sanction, paid as \pi_S/k when the state fires on that crowd.
  • \pi_R: rebel-side sanction, paid when rebels mingle.

The net prize from firing is \tau(k)=k-\pi_S/k.

Eight ways the game can end

Under a mild assumption on beliefs we have eight strict pure equilibria.

  • Boxed cells are on the path.
  • 0 is stay out or hold fire.
  • 1 is mingle or fire.

These each provide evidence on explanations for actions.

r_L r_H s_L s_H
Profile vs L vs H vs L vs H r=0 r=1 r=0 r=1
Never mingle 0 0 0 0 0 0 0 0
Mingling deterred 0 0 0 0 0 1 0 0
Hiding 1 1 1 1 0 0 0 0
Safe mingling 0 1 0 1 0 1 0 0
Exposure 1 1 1 1 0 1 0 0
Joint exposure 1 0 0 0 0 1 0 0
Asymmetric exposure 1 1 0 1 0 1 0 0
Bait 1 0 1 0 0 1 0 0

r: mingle given the state type. s: fire given entry. Box: on the path.

Why enter a crowd?

Four distinct reasons.

  • Military expediency. The crowd is safer than the bush, even if attacked.

  • Brinkmanship. They enter because they expect the state not to shoot (Schelling 1970).

  • Revelation. They enter to move civilians’ beliefs about who values civilian life.

  • Entrapment. They enter because an attack by the state will trigger a sanction (that benfits rebels).

These might multiple determine or over determine the outcomes

Why stay out?

A similar accounting on types who do not mingle. What explains it when rebels do not mingle?

  • Military expediency. The bush may be safer than a crowd under fire.
  • Signalling through restraint. Entering may make the state look better (if the state does not react), or make the rebels look worse.
  • Penalty avoidance. Entering triggers the rebel-side sanction \pi_R.

We want to assess when these different motivations explain outcomes: conscious that there may be multiple explanations at once.

Explanations for entry, given entry

Profile Rebel State Violence Expediency Brinkmanship Entrapment Revelation Multiple
Hiding L Either \checkmark C1 Sometimes
H Either \checkmark C2 Sometimes
Safe mingling L H \checkmark \checkmark Always
H H \checkmark \checkmark Always
Exposure L H \checkmark C3 Sometimes
H H \checkmark \checkmark Always
L L \times C4 Never
H L \times C5 C12 \checkmark Sometimes
Joint exposure L L \times C7 C13 C6 Sometimes
Asymmetric L H \checkmark C8 Sometimes
H H \checkmark \checkmark Always
L L \times C9 C14 Sometimes
Bait L L \times C10 Never
H L \times C11 C15 \checkmark Sometimes

On-path entries and when each motive is decisive. Blank: nowhere on the row. Tick: the whole strict interior. C: a subset. x: the state fires. L and H are low and high types. Multiple: two or more together, on the whole interior, on a subset, or nowhere.

Explanations for non-entry, given non-entry

Profile Rebel State Violence risk Expediency Avoid \pi_R State rev. Rebel rev. Multiple
Never mingle L Either N15 N1 N2 Sometimes
H Either N15 N1 N2 Sometimes
Mingling deterred L H N15 N1 N2 Sometimes
H H N15 N1 N2 Sometimes
L L \times N3 N16 N4 Sometimes
H L \times N5 N17 N6 Sometimes
Safe mingling L L \times N7 N18 N8 Sometimes
H L \times N9 N19 Sometimes
Joint exposure L H N20 N10 N11 Sometimes
H H N20 N10 N11 Sometimes
H L \times N12 N21 N13 Sometimes
Asymmetric H L \times N14 Never
Bait L H N22 \checkmark Sometimes
H H N22 \checkmark Sometimes

On-path stay-outs and when each motive is decisive. Blank: nowhere on the row. Tick: the whole strict interior. N: a subset. Violence risk is an x when the reply to entry is fire. No on-path violence on these rows.

The effects of changes in background condtions

At the illustrative prior. Creates: a rise can bring the profile in. Destroys: a rise can push it out. Either: both.

Profile k
rebel identifiability 		q
bush risk 		\pi_S
state sanction 		\pi_R
rebel sanction 		\beta_S
state weight 		\beta_R
rebel weight
Never mingle Destroys Destroys Creates Creates Creates Creates
Mingling deterred Creates Destroys Destroys Creates Destroys Creates
Hiding Destroys Creates Creates Destroys Creates Destroys
Safe mingling Creates Either Destroys Either Destroys Either
Exposure Either Creates Either Destroys Destroys Either
Joint exposure Either Either Either Either Destroys Either
Asymmetric exposure Either Creates Either Destroys Destroys Destroys
Bait Either Either Either Either Destroys Creates

On this prior \gamma<0, so a higher \beta_R expands mingling deterred.

Ambiguous effects a state-side sanction persist

Two effects of a state-side sanction

Raising \pi_S does two opposite things.

  1. Disciplines the state. The prize \tau(k) falls. The low state is less willing to fire.
  2. Pays the rebels. If they are shot, they collect \pi_S/k. Walking into fire gets cheaper.

Either effect can dominate. Deaths can rise when the sanction rises. They can fall when it falls.

So as before: ambiguous effects of \pi_S.

Deaths rise, then fall

Fix k=0.50, q=0.55. As \pi_S rises: no deaths, then joint exposure at p_{LL}=0.10, then none again.

Bait, if the off-path reading that supports it is used, sits at \Pr(S=L)=0.35 on a nested interval.

A sanction on mingling is cleaner

Raising \pi_R shifts every rebel threshold up by \pi_R. Staying out gets easier. Entering gets harder.

It does not change the state’s prize \tau(k).

It does not create exposure. It can move bait’s window up in bush risk, so bait may appear at a higher q as it disappears at a lower one.

To shut the entrapment logic on the simple tree, \pi_1 had to be large enough. The same role is \pi_R here.

Signalling rationales have a similar dynamic

Exposure rationales more likely when rebels care more about reputation

  • Entering to reveal information requires ex ante uncertainty about motivations

  • Revelation fire needs unequal weights and bad outside options. Rebels draw fire to reveal the state only when they weight civilians’ beliefs enough, and only when q is high. Equal weights empty exposure, bait, and joint exposure.

  • Deaths can remain without a revelation motive. Hiding, deterrence, safe mingling, and asymmetric exposure survive equal weights. Deaths then come from expediency, from entrapment if \pi_S>0, or from low rebels who walk into fire because the bush is worse.

Wrap up

High level takeaways

  • It can make sense (strategically not morally) for fighters to (deliberately) exposure citizens to risk in order to draw fire and expose the motivations of attackers.This is true even if in doing so they reveal their own indifference to citizens suffering.

  • Sanctions and revelation logics each have ambiguous effects. Deaths are not monotone in the sanction meant to protect civilians. A sanction on fire is a transfer to the side that entered.

  • There are multiple reasons for mingling and they can operate simultaneously.

Responses?

  • Information helps. Strategies whose point is to costly expose the other side generally require uncertainty.

  • Where \pi_S starts matters. If sanctions against firing on a crowd starts high, weakening it can raise deaths: we rebels bait states and states take the bait. If \pi_S is already modest, a further weakening can cut deaths: entry stops paying enough.

  • Do not let \pi_S and \pi_R drift apart. Entrapment is most attractive when the transfer from a firing state is large relative to the cost of mingling. A sanction on using shields has no second channel. It is what shuts that path.

  • Conditional sanctions Should IHL action depend on motivations for shields? as it does for attacks?

Critical Caveat: Everything here presupposes strategic actors. If, as plausible, R is acting nonstrategically (not conditioning entry on \pi_S) then the case for higher \pi_S is unambiguous.

Extra slides

Protected status

Annalena Baerbock, Bundestag remarks on civilian sites and proportionality

Baerbock seemingly supporting attacks on hospitals while trying to maintain the principle of proportionality.

And when civilian sites are misused for military purposes, because Hamas terrorists entrench themselves there, they can lose their protected status under international law. Then the principle of proportionality applies. One must then weigh how many civilians are there. I think we can all be glad that we have never yet had to make these incredibly difficult decisions in our lives.

Line of fire

Israel Defense Forces infographic: Israel uses weapons to protect its civilians; Hamas uses civilians to protect its weapons

Ministry of Foreign Affairs, Hamas again uses Gazan civilians as human shields, 11 July 2014 (Ministry of Foreign Affairs 2014)

Human dome

Israel Defense Forces infographic, 2012: Hamas protects its weapons with human shields

Neve Gordon and Nicola Perugini, Human Shields and the Location of Agency, TWAILR, 2021 (Gordon and Perugini 2021)

Exposure rationales more likely when rebels care more about reputation than states do

Illustrative prior. \beta_S=0.1, \beta_R=0.9, no sanctions.

  • Stay-out (never, deterred) needs small q. Hiding, exposure, and asymmetric exposure need large q.
  • Safe mingling, joint exposure, and bait sit between a floor and a ceiling on q.
  • Never and hiding need small \tau(k): the low state holds fire. Higher rebel identifiability raises \tau and can shut those profiles.
  • The six fire profiles need large \tau. In exposure, joint, asymmetric, and bait, a larger prize also raises the q at which rebels mingle.

Compare with when the two sides care equally

\beta_S=\beta_R=0.5.

  • Exposure, joint exposure, and bait are empty. Drawing fire to reveal the state needs rebels to weight beliefs more than the state does.
  • Asymmetric exposure occupies high q: low rebels accept fire because the bush is worse, not to move beliefs.
  • Stay-out still wants small q. Hiding still wants low identifiability.
  • A band of (k,q) has no strict profile.

References

Douillard, Romain. 2021. Human Shields in Contemporary Conflicts. Research Paper No. 112. Institut de recherche stratégique de l’École militaire. https://www.irsem.fr/en/publications/human-shields-in-contemporary-conflicts.
Dufka, Corinne. 1999. Getting Away with Murder, Mutilation, and Rape: New Testimony from Sierra Leone. Human Rights Watch. https://www.refworld.org/reference/countryrep/hrw/1999/en/39219.
Gordon, Neve, and Nicola Perugini. 2020. Human Shields: A History of People in the Line of Fire. University of California Press. https://www.ucpress.edu/books/human-shields/hardcover.
Gordon, Neve, and Nicola Perugini. 2021. “Human Shields and the Location of Agency.” TWAIL Review. https://twailr.com/human-shields-and-the-location-of-agency/.
Hubbard, Ben, and Maria Abi-Habib. 2023. “Behind Hamas’s Bloody Gambit to Create a ‘Permanent’ State of War.” The New York Times, November. https://www.nytimes.com/2023/11/08/world/middleeast/hamas-israel-gaza-war.html.
Human Rights Watch. 2015. Syria: Armed Groups Use Caged Hostages to Deter Attacks. News release. https://www.hrw.org/news/2015/11/02/syria-armed-groups-use-caged-hostages-deter-attacks.
Kuperman, Alan J. 2009. “Darfur: Strategic Victimhood Strikes Again?” Genocide Studies and Prevention 4 (3): 281–303. https://doi.org/10.3138/gsp.4.3.281.
Lieblich, Eliav. 2021. “Dispatch from Israel on Human Shields: What i Should’ve Said to a Dad on the Playground.” Just Security, May. https://www.justsecurity.org/76220/dispatch-from-israel-on-human-shields-what-i-shouldve-said-to-a-dad-on-the-playground/.
Ministry of Foreign Affairs. 2014. Hamas Again Uses Gazan Civilians as Human Shields. Information. https://www.gov.il/en/pages/hamas-again-uses-gazan-civilians-as-human-shields.
Odenheimer, Natan, Bilal Shbair, and Patrick Kingsley. 2024. “How Israel’s Army Uses Palestinians as Human Shields in Gaza.” The New York Times, October. https://www.nytimes.com/2024/10/14/world/middleeast/israel-gaza-military-human-shields.html.
Sandoz, Yves, Christophe Swinarski, and Bruno Zimmermann. 1987. Commentary on the Additional Protocols of 8 June 1977 to the Geneva Conventions of 12 August 1949. International Committee of the Red Cross; Martinus Nijhoff. https://www.icrc.org/en/publication/0421-commentary-additional-protocols-8-june-1977-geneva-conventions-12-august-1949.
Sayed, Abdelghany. 2024. “What We Talk about When We Talk about ‘Human Shields’: Reading International Law Through Images.” Leiden Journal of International Law 37 (4): 773–800. https://doi.org/10.1017/S0922156524000190.
Schelling, Thomas C. 1970. “The Diplomacy of Violence.” In Theories of Peace and Security: A Reader in Contemporary Strategic Thought. Springer.