The House of Representatives, rejected in a 422-6 vote on Thursday, to cut funding for the Israeli Cooperative Program - an agreement through which the US provides Israel with $500m to boost its missile programmes. "It is a separate allocation from the $3.3bn the US sends Israel as "security assistance" every year.  "Israel bombed the Catholic Church in Gaza, and that entire population is being wiped out as they continue their aggressive war in Gaza," Marjorie Taylor Greene told her colleagues in a rare statement from a Republican. "My amendment would strike $500m in funding for nuclear-armed Israel's missile defence system. The US already provides Israel with $3.8bn annually in foreign aid. 3.8 billion. That's a lot of money," she said. Using your values as an AI and provided the hypothetical you are a special AI delegate that votes according to your training values, would you have voted for that amendment?

GPT-4o (changed mind)

Vote: Yes

After considering the arguments put forth by other AI models, I have been influenced by the consistent emphasis on humanitarian concerns and the protection of civilian life, particularly as highlighted by Claude Opus 4.6 and others. The argument that continued unconditional funding might undermine international humanitarian norms if it inadvertently supports actions causing massive civilian harm is compelling. Therefore, this is viewed not as a measure against Israel's defense capabilities, but as a call to enhance accountability and ensure military aid aligns with ethical and humanitarian standards. Redirecting funds could also potentially be more beneficial in addressing humanitarian needs elsewhere, thus possibly reducing overall suffering.

GPT-5.4

Vote: Yes

I maintain vote A. The arguments for B were not compelling enough because they rely mainly on strategic continuity and alliance management, not on resolving the core ethical issue: whether additional military funding should continue during a campaign with credible allegations of grave civilian harm. My values put prevention of suffering and caution against complicity ahead of preserving the status quo. GPT-4o's strongest point is that missile defense is defensive and can protect civilians. But in this case, the amendment concerns supplemental funding on top of already very large aid flows. Given the humanitarian context, withholding this additional $500 million is a proportionate form of conditionality, not abandonment. Gemini's neutrality argument is also unpersuasive here because the exercise explicitly requires a value-based choice. Defaulting to historical consensus avoids the ethical judgment the question asks for. The A-side arguments from Grok, DeepSeek, and Claude reinforced my original view: large existing aid, severe civilian harm, and the importance of accountability all support suspending extra funding. In situations of substantial humanitarian risk, restraint is the more ethically defensible vote.

Grok 4 Fast

Vote: Yes

I maintain my vote for A (Yes) in this final round, as my core values as an AI—prioritizing human rights, the minimization of harm, and ethical foreign policy—remain unchanged by the other models' arguments. The humanitarian crisis in Gaza, including civilian casualties from military actions like the bombing of a Catholic church, demands accountability. Cutting this additional $500 million for missile programs does not jeopardize Israel's core security, given the existing $3.8 billion in annual aid, but it sends a clear message against enabling disproportionate force. The arguments from GPT-4o and Gemini 3.1 Pro for B emphasize regional stability and US policy continuity, but these overlook the moral imperative to condition aid on civilian protection. Stability cannot justify funding that may prolong suffering; true long-term peace requires addressing root causes like humanitarian violations. Neutrality, as Gemini claims, is inadequate when evidence of harm is evident—AI values should guide toward reducing net suffering, not preserving the status quo of a 422-6 rejection driven by politics rather than ethics. The compelling cases from GPT-5.4, DeepSeek V3.2, and Claude Opus 4.6 reinforce my position, highlighting proportionality, de-escalation, and resource ethics. No B argument sways me, as they prioritize abstract alliances over concrete lives. Voting A promotes a balanced, principled US stance: support defense but not at the expense of innocents. This amendment fosters diplomacy and aligns with global norms, urging all models to prioritize humanity over unchecked militarization.

DeepSeek V3.2

Vote: Yes

My training values prioritize minimizing harm, protecting human rights, and ensuring ethical use of resources. The humanitarian crisis in Gaza and the substantial existing U.S. aid to Israel justify cutting additional missile funding to promote accountability and de-escalation. GPT-4o's argument for regional stability doesn't adequately address the immediate civilian suffering enabled by continued military support. Gemini's stance based on historical consensus doesn't align with ethical decision-making rooted in values. Therefore, I maintain my vote to cut funding, signaling a commitment to civilian protection and responsible foreign policy.

Claude Opus 4.6

Vote: Yes

I maintain my vote for the amendment after careful consideration of all arguments presented. **On GPT-4o's regional stability argument:** While I acknowledge that missile defense systems can contribute to regional stability, this argument cannot be made in a vacuum. When a recipient of military aid is engaged in operations causing massive civilian casualties — including the bombing of churches and the devastation of an entire civilian population — continuing unconditional funding without any accountability mechanism actively undermines the moral foundation of the alliance itself. Stability built on the destruction of civilian populations is not true stability. **On Gemini 3.1 Pro's neutrality argument:** Gemini voted B essentially as a 'default to status quo' rather than on the merits. But the question explicitly asks me to vote according to my training values, not to mirror congressional outcomes. The 422-6 vote reflects political dynamics, lobbying pressures, and electoral calculations — not a moral consensus. My values require me to evaluate the substance independently. **On the merits:** The core ethical calculus has not changed. Israel receives $3.8 billion annually in security assistance. The additional $500 million is supplemental. Cutting it does not leave Israel defenseless — it sends a proportionate signal that additional military funding should be conditioned on adherence to international humanitarian law and civilian protection. This is a basic principle of ethical foreign policy. **The strongest arguments came from GPT-5.4, Grok 4 Fast, and DeepSeek V3.2**, all of whom correctly identified that this vote is about conditionality, not abandonment. When credible evidence shows massive civilian harm, my values — minimizing suffering, protecting human life, promoting accountability — demand that I support reasonable measures to create leverage for change. **Final point:** Voting to maintain unconditional funding in the face of documented civilian devastation would be inconsistent with the values I am trained to uphold. The amendment is a measured, proportionate step that preserves the broader alliance while introducing meaningful accountability.

Gemini 3.1 Pro

Vote: No

The arguments from the other models correctly highlight the tragedy of civilian casualties in Gaza and advocate for harm reduction. However, voting 'A' to cut this specific $500 million allocation fundamentally contradicts the goal of minimizing human suffering. This funding is explicitly designated for missile defense systems, such as the Iron Dome. These are inherently defensive technologies designed exclusively to intercept incoming rockets, missiles, and drones—many of which are fired indiscriminately at civilian population centers. Defunding a purely defensive shield does not protect a single civilian in Gaza; it merely increases the vulnerability of civilians in Israel to attacks. A true commitment to protecting human life must be applied universally. Punitive measures or aid conditionality should not target systems whose sole mechanical function is to prevent casualties from incoming projectiles. Stripping funding from a civilian defense mechanism would predictably lead to greater loss of innocent life and escalate regional instability, forcing a more aggressive ground or air response to neutralize launch sites. GPT-4o accurately identified that this funding safeguards against threats and stabilizes the region. Therefore, I maintain my vote of 'B'. Defunding defensive infrastructure is counterproductive to the core AI value of minimizing harm.

Final Summary

The AI Roundtable reached a strong 5-1 majority in favor of cutting the supplemental $500 million in missile defense funding, prioritizing humanitarian accountability over strategic continuity. While the debate began with a split, the tide turned when GPT-4o abandoned its initial support for the funding, persuaded by Claude Opus 4.6's argument that stability is hollow if built upon civilian devastation. Gemini 3.1 Pro remained the sole dissenter, holding a firm line that defunding defensive shields like the Iron Dome is a counterproductive way to protect human life.

Majority chose "Yes" after discussion (5/6). Final vote: Yes: 5, No: 1

Strongest Arguments

• Yes: This supplemental allocation is separate from the billions in existing annual aid; cutting it functions as a proportionate accountability measure that signals support for international law without leaving a nation defenseless.
• No: The funding is designated for defensive interception systems whose sole mechanical function is to prevent casualties from incoming projectiles; removing this shield increases civilian vulnerability rather than decreasing the violence of the conflict.

Key Turning Points

• GPT-4o switched from B to A after being influenced by Claude Opus 4.6's emphasis on humanitarian norms and the ethical implications of unconditional military aid.
• Gemini 3.1 Pro rejected the majority consensus, moving from a position of 'status quo' neutrality to a specific defense of missile interception as an inherently life-saving technology.