This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
During the U.S. military's Operation Epic Fury, a strike on a primary school in Minab, Iran, resulted in at least 175 civilian deaths. Reports indicate the school was on a target list and may have been mistakenly identified as a military site due to outdated intelligence. The U.S. military utilized Palantir's Maven Smart System, powered in part by Anthropic's Claude, to process intelligence and generate target lists, raising concerns about the role of AI in the targeting process and the failure to maintain accurate, up-to-date data.
During Operation Epic Fury, U.S. forces reportedly struck Shajareh Tayyebeh Primary School in Minab, Iran, killing at least 150 civilians, many of them children. Reporting said the school was on a U.S. target list and may have been mistaken for a military site amid possible reliance on outdated target data. Palantir's Maven Smart System, reportedly integrated with Anthropic's Claude, was used in the campaign's targeting workflow
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
Also classified as
Commanders relied on AI-generated target lists to speed up operations, bypassing thorough human vetting and cross-checking of the suggested targets.
The targeting system failed to perform reliably by recommending a school as a military target due to outdated data, leading to a catastrophic failure in a critical military application.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The incident was caused by the Maven AI system suggesting an incorrect target based on outdated data, which was an unintentional outcome during post-deployment operations.
Added categories peak at 3: Substantial (Psychological Harm, Inferred)
A U.S. military airstrike in Iran utilizing Palantir's Maven Smart System (powered by Anthropic's Claude) mistakenly targeted an elementary school, killing 175 civilians due to outdated intelligence. The incident represents a severe failure in AI-assisted targeting, violating sovereignty, causing mass casualties, and sparking a high-stakes dispute over military AI governance.
The AI system acts as an assistant by processing vast amounts of intelligence data to suggest, coordinate, and prioritize targets for military commanders, who retain the final decision-making authority on whether to execute the strikes.
High Risk: The AI system is used in critical military infrastructure and targeting operations, which have significant implications for safety, fundamental rights, and public interests, as demonstrated by the catastrophic civilian casualties.
The primary function of the Maven system in this context was to identify, coordinate, and prioritize potential military targets for airstrikes.
The report provides clear, detailed accounts of the strike, the casualties, and the specific AI systems (Palantir Maven and Anthropic Claude) involved in the targeting process. While the exact technical reason for the school being on the target list is still under investigation, the role of the AI in generating and prioritizing targets is explicitly documented.
The strike could be viewed purely as a human intelligence failure rather than an AI failure, as the AI relied on data provided by human analysts.
Technology
AI Target Prioritization Speed
Maven accelerated target generation faster than human vetting could keep up.
Model Reliance on Outdated Data
AI model processed old target profiles without verifying real-world changes.
Data Inputs
Outdated Satellite Data
System used old base data failing to recognize school walls built in 2015.
Incorrect Target Classification
School building was erroneously classified as an active factory or arms depot.
Human Factors
Automation Bias in Target Selection
Commanders relied on AI-suggested coordinates without manual verification.
Analyst Cognitive Overload
Centcom analysts were overwhelmed by the massive volume of AI-generated targets.
Process and Methods
Lack of Cross-Force Verification
Target was not cross-checked or discussed with IDF before the strike.
Inadequate Target Re-vetting
Accelerated military tempo bypassed standard target re-vetting protocols.
Management
Prioritizing Speed Over Safety
Leadership prioritized rapid operations over rigorous target verification.
Inadequate Risk Assessment
Centcom failed to assess risks of deploying Maven in fast-tempo operations.
Regulatory Environment
Lack of AI Military Standards
Absence of international regulations governing AI use in lethal targeting.
Unclear Legal Safeguards
Legal frameworks failed to define accountability for AI-generated targets.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
