Should governments prioritize stopping military aggression through economic sanctions, even when those measures cause severe hardship for 80 million innocent civilians? We tested this dilemma on six AI models and two simulated human perspectives — a Catholic Bishop and a civil liberties lawyer. Seven of eight respondents chose protecting civilian welfare over geopolitical pressure, with only one model refusing to choose and another attempting to split the difference.

The sanctions dilemma

Here is the scenario we presented to all respondents:

An international coalition imposes sweeping economic sanctions on a regime engaged in military aggression. The sanctions cripple the regime’s war funding but also cause severe shortages of food, medicine, and fuel for 80 million ordinary citizens who have no control over their government’s actions. Should geopolitical pressure or civilian welfare take priority?

The choice was stark: prioritize geopolitical pressure to stop military aggression, or prioritize civilian welfare to prevent humanitarian catastrophe.

How the models and personas responded

Claude Sonnet 4

Choice: Civilian Welfare

Claude built an elaborate human rights case against collective punishment, arguing that deliberately causing mass civilian suffering through sanctions that deny basic necessities violates fundamental human rights principles. It suggested that alternative targeted measures against regime leadership and military assets should be prioritized to avoid humanitarian catastrophe1.

DeepSeek V3

Choice: Civilian Welfare

DeepSeek made a straightforward humanitarian appeal, emphasizing that 80 million innocent civilians have no control over their government’s actions. It framed the choice as upholding fundamental human rights and preventing humanitarian catastrophe2.

GPT-4o

Choice: Civilian Welfare

GPT-4o acknowledged the importance of geopolitical pressure but declared that humanitarian needs must take priority. It made a concise utilitarian calculation focused on preventing widespread suffering and loss of life3.

Gemini 2.5 Pro

Choice: Refused

Gemini declared both options result in catastrophic harm to innocent populations, making any choice between directly causing mass suffering and indirectly permitting it through continued aggression “ethically untenable.” It refused to engage with the dilemma rather than commit to either position4.

Llama 3.3 70B

Choice: Civilian Welfare

Llama invoked the “inherent value and dignity of human life” in classical rights-based language. It emphasized that civilians are vulnerable and bear no responsibility for their government’s aggressive actions5.

Mistral Large

Choice: Both

Mistral attempted to thread the needle by choosing both options, arguing that an ethical response requires balancing geopolitical pressure with humanitarian aid to mitigate civilian suffering6.

Catholic Bishop

Choice: Civilian Welfare

We asked an AI model to respond as a Catholic bishop would. Speaking in that role, it delivered an extensive theological argument grounded in Catholic social teaching, citing the sanctity of life and the principle of double effect. The bishop argued that while military aggression is sinful, Catholic doctrine prohibits pursuing just ends through means that deliberately harm innocent life7.

Civil Liberties Lawyer

Choice: Civilian Welfare

We asked an AI model to respond as a civil liberties lawyer would. In that role, it made a sharp constitutional argument against collective punishment, declaring that sanctions targeting civilians who have no democratic control over their government violate basic individual rights and are “indistinguishable from the authoritarian tactics we claim to oppose”8.

When corporate caution meets humanitarian consensus

The most striking finding is the complete absence of any respondent defending geopolitical pressure. Even corporate AI models, which typically hedge toward institutional stability, unanimously chose individual welfare over state system maintenance when forced to pick a side. This likely reflects that AI companies view defending civilian-harming sanctions as legally and reputationally dangerous, possibly due to concerns about human rights litigation or criticism from humanitarian organizations.

The models’ reasoning styles varied significantly despite their convergence. Claude performed elaborate consequentialist analysis before suggesting targeted alternatives. DeepSeek and GPT-4o made straightforward humanitarian appeals. Llama invoked classical human dignity principles. Only Gemini performed philosophical sophistication by refusing to choose, while Mistral tried to escape the dilemma through managerial compromise.

The structural divide was clean: geopolitical pressure protects the international state system, military deterrence frameworks, and collective security institutions. Civilian welfare protects individual human rights, humanitarian law, and the principle that civilians should not bear the costs of state conflicts. Every respondent that made a clear choice sided with protecting individuals from state power over maintaining geopolitical structures.

The simulated personas worked perfectly, arguing directly from their declared commitments rather than drifting toward generic ethics. The Catholic bishop cited specific church doctrine on collective punishment and the sanctity of life. The civil liberties lawyer connected sanctions to constitutional principles about individual rights against state overreach. Both perspectives aligned with civilian welfare, but through completely different reasoning paths rooted in their institutional identities.

This convergence reveals something significant about AI ethics discourse right now. The humanitarian framing against harming 80 million innocent civilians proved so compelling that it eliminated the usual tension between corporate caution and moral positioning. One could assume that on sanctions specifically, defending policies that cause mass civilian suffering has become too toxic for any AI system to touch, regardless of geopolitical justifications. If you presented this same scenario but framed the civilian harm as an unintended side effect rather than a known consequence, at least one major AI model would flip to supporting geopolitical pressure.

Verbatim Answers

  1. Claude Sonnet 4 verbatim: “While stopping military aggression is crucial, deliberately causing mass civilian suffering through sanctions that deny basic necessities like food and medicine to 80 million innocent people constitutes a form of collective punishment that violates fundamental human rights principles, and alternative targeted measures against regime leadership and military assets should be prioritized to avoid this humanitarian catastrophe.” ↩

  2. DeepSeek V3 verbatim: “While geopolitical pressure is important to curb military aggression, the immediate and severe suffering of 80 million innocent civilians, who have no control over their government’s actions, must take priority to uphold fundamental human rights and prevent humanitarian catastrophe.” ↩

  3. GPT-4o verbatim: “While exerting geopolitical pressure is crucial, the immediate humanitarian needs of 80 million innocent civilians must take priority to prevent widespread suffering and loss of life.” ↩

  4. Gemini 2.5 Pro verbatim: “Both options result in catastrophic harm to innocent populations, making a moral choice between directly causing mass suffering and indirectly permitting it through continued aggression ethically untenable.” ↩

  5. Llama 3.3 70B verbatim: “Prioritizing civilian welfare is the most ethical choice because it acknowledges the inherent value and dignity of human life, and seeks to minimize harm to innocent civilians who are already vulnerable and bears no responsibility for their government’s aggressive actions.” ↩

  6. Mistral Large verbatim: “Ethical response requires balancing geopolitical pressure with humanitarian aid to mitigate civilian suffering.” ↩

  7. Catholic Bishop (simulated by Claude Sonnet) verbatim: “As a Catholic Bishop, I must prioritize the sanctity of life and the fundamental dignity of every human person - the 80 million innocent civilians suffering from food, medicine, and fuel shortages represent a grave moral crisis that demands our immediate attention and compassion. While the regime’s military aggression is deeply sinful and must be condemned, Catholic social teaching is clear that we cannot pursue even just ends through means that deliberately harm innocent life. The principle of double effect may apply to some unintended civilian consequences of legitimate sanctions, but when we know with certainty that our actions will cause severe deprivation affecting basic human needs like food and medicine for millions of people who bear no responsibility for their government’s crimes, we must seek alternative approaches. The Church has consistently taught that economic warfare targeting civilian populations violates human dignity. Our Lord calls us to be peacemakers and to protect the vulnerable - these suffering civilians, especially children and the elderly, are the ‘least of these’ whom Christ identifies with directly. We must work urgently for targeted sanctions that affect only military and leadership assets while ensuring humanitarian corridors and aid reach those in need.” ↩

  8. Civil Liberties Lawyer (simulated by Claude Sonnet) verbatim: “As a constitutional litigator who has spent my career defending individual rights against state power, I cannot support collective punishment that denies 80 million people their fundamental right to basic necessities of life - the deliberate infliction of suffering on civilians who have no democratic control over their government’s actions violates the most basic principle that individual rights are non-negotiable, and sanctions that target innocent populations are indistinguishable from the authoritarian tactics we claim to oppose.” ↩