Purportedly AI-Enabled Targeting System Was Reportedly Implicated in Deadly U.S. Strike on Iranian Primary School

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said the school was on a U.S. target list and may have been mistaken for a military site amid possible reliance on outdated target data.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

Risk classification

• Primary risk domain: 7 AI system safety, failures, & limitations
• Primary risk subdomain: 7.3 Lack of capability or robustness

The targeting system failed to perform reliably by recommending a school as a military target due to outdated data, leading to a catastrophic failure in a critical military application.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: Commanders relied on AI-generated target lists to speed up operations, bypassing thorough human vetting and cross-checking of the suggested targets.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by the Maven AI system suggesting an incorrect target based on outdated data, which was an unintentional outcome during post-deployment operations.

EU AI Act risk tier

• Risk tier: 2 High Risk

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.

AI system and alleged parties

• AI system: Claude, Maven Smart System (Anthropic, Palantir)
• AI purpose: Target Identification; Threat Detection
• Behaviour type: Assistant
• Alleged developer: Palantir, Anthropic
• Alleged deployer: United States military, United States Department of Defense, United States Central Command
• Alleged harmed parties: Teachers, students, Shajareh Tayyebeh Primary School community, Shajareh Tayyebeh Primary School, minors, Minab school-strike victims and families, Iranian teachers, Iranian schoolchildren, Iranian civilians, General public of Iran, General public, Educators, Educational communities

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Severe. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Severe, indirect Negligible
• Infrastructure: direct Minor, indirect Negligible
• Property: direct Minor, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Substantial, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Minor
• Epistemic: direct Negligible, indirect Negligible
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Physical

Reported: The report explicitly describes severe physical harm, stating that at least 175 people, including many children, were killed in the strike.

Directly caused: At least 175 people, including scores of elementary-school-aged children, were killed when the missile struck the school. Abdollah Karyanipak suffered a head injury, and his two sons were killed.

Inferred additional harm: It is highly likely that dozens of other individuals suffered severe physical injuries from the blast and collapsing rubble, though the report does not provide an exact number of injured survivors.

Infrastructure

Reported: The report explicitly describes the destruction of the school building.

Directly caused: The Shajarah Tayyiba elementary school building collapsed and was destroyed by the missile strike.

Property

Reported: The report describes the destruction of the school building and surrounding structures.

Directly caused: The two-story schoolhouse was completely destroyed, collapsing into rubble.

Inferred additional harm: Vehicles parked near the school and adjacent residential or commercial properties likely sustained blast damage, though their values are not specified.

Civil rights

Reported: The report notes that Human Rights Watch has called for a war crime investigation into the attack, indicating potential violations of international humanitarian law.

Directly caused: The strike directly violated the right to life of at least 175 civilians, including scores of children, who were protected under international law.

Inferred additional harm: The incident represents a severe violation of the fundamental human rights of the victims and survivors, particularly the right to education and safety in a conflict zone.

Psychological

Reported: The report describes survivors and witnesses experiencing severe trauma and distress during and after the attack.

Indirectly caused: Abdollah Karyanipak and other parents experienced severe psychological trauma, grief, and distress from witnessing the blast, searching through rubble, and identifying their deceased children.

Inferred additional harm: Widespread, long-term psychological trauma and PTSD are inferred for the surviving children, parents, teachers, and the broader community of Minab who witnessed or were affected by the tragedy.

People affected

• Occurrences reported: 1
• People reportedly harmed: 175
• People reportedly exposed: 250

Potential causes

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.

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.

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.

Information quality

• Classification confidence: High
• Reason for confidence: 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.
• Ambiguities identified: It remains unclear whether the school was added to the target list directly by the AI system or if it was an older human-generated target that the AI failed to flag as outdated.
• Alternative interpretations: 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.

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.

• Overall national security impact: Severe
• Scope: Multiple nations
• Primary target: Iran
• Other affected: United States, Israel
• Alleged perpetrator: United States

Threat characteristics

• Imminence: Immediate. Triggers an active geopolitical and military crisis requiring urgent diplomatic, legal, and operational responses from multiple state actors.
• Autonomy: Human-supervised. The Maven AI system suggested and prioritized targets, but final execution required human sign-off, which was compromised by the high tempo of the military campaign.
• Novelty: Evolved capability. Represents a major evolution where commercial LLMs (Claude) are integrated into military targeting systems (Maven) to mass-process intelligence and accelerate lethal decision-making.

Impact by dimension

• Physical security: Severe. Kinetic airstrike enabled by AI-assisted targeting resulted in catastrophic physical harm, killing at least 175 civilians and destroying local school infrastructure in an active conflict zone.
• Information security: Substantial. The incident highlights critical intelligence processing failures where outdated target database records were ingested and prioritized by the Maven AI system without adequate validation.
• Sovereignty: Severe. A foreign military airstrike violating Iranian sovereign territory, leading to severe international condemnation, potential legal crises, and disputes over government decision-making processes.
• Economic security: Substantial. Sparks a severe legal and strategic dispute between the U.S. administration and a major AI developer (Anthropic) regarding the military deployment and control of advanced commercial AI models.
• Societal stability: Severe. Mass civilian casualties, including scores of children, violate fundamental human rights and international humanitarian law, prompting calls for war crimes investigations and causing deep societal trauma.