The use of foundational AI is sweeping across the current middle-east conflict, with the US engaging Claude, the artificial intelligence model built by startup Anthropic, through defence contractor Palantir. The use of AI has reportedly sharpened strikes by the US, raising conversations around AI turning into killing machines and going against pledges and constitutions by Big Tech. Is this an inflection point for AI policy worldwide? Mint explains.
How has the US engaged AI on the battlefield so far?
A report by US security research firm Soufan Center underlined that the US Department of War (formerly Department of Defense) used Anthropic’s Claude, embedded in Palantir’s battlefield intelligence suite Maven Smart System, to identify sharp targets during its strikes on Iran, called Operation Epic Fury. The uses also included simulating battlefield situations and strategies. At the same time, DoW also signed a deal with fellow AI maker OpenAI to use its GPT foundational models in future conflicts. It isn’t clear whether GPT has already been deployed and is in use, as much of wartime tools remains classified.
Did tech firms build AI models specifically aligned for war?
Not as per public disclosures. Mint’s examination of US DoW statements and public documents show that Palantir’s battlefield intelligence system was the primary interface for using Claude, which was commercially used by the latter for orchestrating attacks that have so far killed over 1,200 individuals in Iran. But Anthropic’s 84-page ‘constitution’ for Claude clearly states even now that it will not be used for “actions that are inappropriately dangerous or harmful.” OpenAI chief Sam Altman, in a 27 February statement, said the company’s DoW deal includes “human responsibility for the use of force, including for autonomous weapon systems.”
Is this the first use of AI on battlefields?
Not really. Weapon systems such as guided missiles and drones have been using machine learning, object recognition, computer vision and more for remote strikes. But criticism has surfaced around AI use in battlefields because of the ability of models to recognize increasingly sophisticated patterns at a rapid pace, and also identify strategic weaknesses in the real world. Massive compute availability further accelerates the ability of nations to carry out surgical strikes on rivals.
What does this mean for the future of AI policies?
Geopolitics consultants have called for audits of tech companies building AI models, and possible globally binding policy agreements, akin to 1968’s nuclear non-proliferation treaty. While OpenAI has claimed it has voluntarily drawn ‘red lines’ to prevent the use of AI in cases of deliberate harm, conflicting reports on how contemporary AI, generative and computational alike, have raised questions around whether nations in leadership positions of AI, such as the US, China and even India, should work toward disarmament clauses for the use of AI in war. This, though, may not prevent the use of AI in defensive battlefield intelligence.
Can AI really become ‘killer machines’, if not controlled?
The possibility is absolutely real. Contemporary AI models are adept at learning visual patterns very fast, and their productive usage can boost agriculture and sustainable land usage. On the other hand, the same pattern recognition can, at least theoretically, identify targets that humans indicate as threats. Such usage has already played out in the ongoing conflict, leading to concerns around the rise of autonomously operational machines that can strike remotely and at will. With vast private funds and increasing revenue generation, any firm building foundational AI can provide the basis for swarms of strike drones and planes, to any country.
Facts Only
The U.S. Department of War (formerly Department of Defense) used Anthropic’s Claude AI model, embedded in Palantir’s Maven Smart System, for target identification during Operation Epic Fury strikes in Iran.
Over 1,200 individuals in Iran have reportedly been killed in attacks orchestrated using Palantir’s system with Claude.
The U.S. Department of War signed a deal with OpenAI to use GPT foundational models in future conflicts; deployment status remains classified.
Anthropic’s 84-page constitution for Claude states it will not be used for "inappropriately dangerous or harmful" actions.
OpenAI’s Sam Altman stated in a February 27 announcement that their DoW deal includes "human responsibility for the use of force, including for autonomous weapon systems."
AI has previously been used in guided missiles and drones via machine learning, object recognition, and computer vision.
Geopolitical consultants advocate for global AI policy agreements, similar to the 1968 Nuclear Non-Proliferation Treaty.
OpenAI claims voluntary "red lines" to prevent AI use in deliberate harm, but conflicting reports exist on AI’s role in warfare.
Contemporary AI models can rapidly recognize visual patterns, theoretically enabling autonomous targeting in military operations.
Executive Summary
Full Take
The strongest version of this narrative highlights a critical inflection point: AI’s dual-use nature is colliding with geopolitical power struggles, exposing the fragility of corporate ethical frameworks when faced with state military demands. The U.S. deployment of Claude and GPT in lethal operations—despite Anthropic’s and OpenAI’s public constitutions—reveals a systemic tension between profit-driven innovation and unaccountable warfare. This isn’t just about "killer robots" but about how AI’s pattern-recognition prowess accelerates the fog of war, enabling strikes with plausible deniability ("the algorithm identified the target"). The call for a nuclear-style treaty is compelling, yet history shows such agreements falter when enforcement lacks teeth or when nations prioritize strategic advantage over collective security.
Patterns detected: **ARC-0024 Ambiguity** (vague framing of "inappropriately dangerous" actions), **ARC-0043 Motte-and-Bailey** (tech firms retreat to ethical constitutions while their tools enable warfare).
Root cause: The paradigm assumes AI can be "controlled" by voluntary corporate guardrails, ignoring how state actors exploit commercial innovation for military ends. This echoes the Cold War’s arms races, where dual-use technologies (e.g., nuclear energy) blurred civilian and combat applications. The unstated assumption? That AI’s benefits (e.g., agricultural optimization) outweigh its risks—yet the same algorithms that map crop yields can map strike zones.
Implications: Human agency erodes when accountability diffuses across AI developers, defense contractors, and policymakers. The costs are borne by civilians in conflict zones, while benefits accrue to states and firms monetizing AI’s dual-use potential. Second-order consequences include normalized autonomous warfare, where "surgical strikes" become a euphemism for algorithmic violence, and a global AI arms race that outpaces governance.
Bridge questions: If tech firms cannot enforce their own ethical constitutions, what mechanisms could? How might AI’s role in warfare reshape international law’s definitions of proportionality and distinction? What would it take for a major power to unilaterally disarm its AI capabilities—and would that even be strategically rational?
Counterstrike scan: A bad-actor playbook would amplify moral panic ("AI is already killing!"), frame the issue as a binary (ban AI or accept doom), and exploit corporate hypocrisy to discredit all AI governance efforts. This article avoids that trap by presenting nuanced tensions—tech firms’ ethical claims vs. military realities—without descending into alarmism. The content doesn’t match a coordinated influence pattern; it’s a legitimate critique of structural gaps in AI policy.
Sentinel — Human
The article shows some stylometric and attribution patterns common in AI-assisted text but retains enough human idiosyncrasies and narrative unevenness to suggest likely human authorship with possible light editorial AI assistance.