Operation Epic Fury generated, across eleven days of AI-enabled strikes against Iran, what is almost certainly the largest and most detailed operational dataset on high-intensity AI-integrated warfare ever assembled. Every strike was processed through the Maven Smart System.
Every target prioritization was generated by algorithms integrating drone feeds, satellite imagery, signals intelligence, and open-source data at machine speed. Every battle damage assessment was processed by the same architecture. The data exists. It is voluminous, operationally consequential, and almost entirely ungoverned by any framework, legal, contractual, or ethical, that addresses what happens to it after the conflict ends.
The ownership question has four distinct dimensions, each with its own legal and strategic implications, and the existing governance architecture addresses none of them adequately. The first is straightforward: who owns the operational data generated by commercial AI systems, Palantir’s Maven Smart System, Anthropic’s Claude before its expulsion, OpenAI’s systems after, when those systems are used in military targeting?
Standard government contracting provisions typically assign intellectual property rights over custom deliverables to the government while allowing contractors to retain rights over underlying technology. But the operational data generated by AI targeting systems during combat operations is neither a custom deliverable nor underlying technology.
It is a new category, call it combat-generated training data, whose ownership, use rights, and security classification status are legally ambiguous.
The Pentagon has signaled its intentions clearly and the implications are significant. MIT Technology Review reported in April 2026 that the Pentagon plans to allow AI companies to train new models on classified military data. This means the operational record of Epic Fury, targeting decisions, strike outcomes, error rates, civilian casualty patterns, may become training material for the next generation of military AI systems.
The companies whose models train on this data will possess the most operationally validated military AI training sets in existence. The strategic value of that data to a competitor who acquired it, through commercial partnership, security breach, or legal process in a jurisdiction where the classifying state has limited control, is not theoretical. It is the kind of data advantage that shapes military AI capability for a generation.
The second dimension is accountability. The Minab school strike, attributed by multiple independent investigations to U.S. Tomahawk missiles, killed over 170 people. No accountability process has been initiated. The operational data that would allow a reconstruction of the targeting decision, what intelligence inputs the system processed, what recommendation it generated, what human review occurred before the strike was authorized, what the time interval was between recommendation and execution, is classified, proprietary, or both.
International humanitarian law requires post-strike assessment of civilian casualty incidents. When the targeting system that generated the strike recommendation is a commercial AI platform operating under a classified government contract, the mechanisms for conducting that assessment do not exist.
Families of victims have no legal recourse. International investigative bodies have no access to the relevant data. The state that conducted the strike has no obligation to disclose the operational record of an AI system it describes as a decision-support tool rather than a targeting system.
The third dimension concerns the states that host military operations conducted by allied forces using AI targeting systems. Pakistan, which hosts U.S. military logistics and intelligence infrastructure, and the Gulf states whose territory hosts U.S. air defense systems operating AI-enabled intercept prioritization, are jurisdictions within which AI-assisted military operations have been conducted without any agreement about what data those operations generate, who owns it, and what legal obligations it creates.
When the IRGC struck AWS data centers in the UAE and attributed the targeting to the presence of U.S. military AI workloads, the UAE had no legal framework that addressed its status as a host of dual-use military AI infrastructure. It had no contractual relationship with the AI companies whose systems were targeted on its territory. It had no claim on the operational data generated within its jurisdictional borders by systems it had not licensed and could not inspect.
The fourth dimension is the competition dynamic. MIT Technology Review documented the Pentagon’s stated intention to train future military AI on the combat data of current operations. This creates a competitive flywheel: the state that conducts the most AI-integrated military operations generates the most operationally validated training data, which produces the most capable military AI systems, which generate more operationally validated data.
States that do not conduct such operations, because they lack the capability, the political will, or because they are on the receiving end of those operations, fall progressively further behind in military AI capability. The data generated by Epic Fury is not merely a record of what happened. It is a strategic asset whose future use will shape the military AI balance of the coming decade, and it is entirely unregulated.
The governance gap here is not abstract. It is the specific question of whether any international framework, arms control, data protection, intellectual property, or international humanitarian law, addresses combat-generated AI training data as a category requiring regulation. The answer, currently, is no. The CCW talks on autonomous weapons have not addressed it.
The GDPR and its equivalents do not apply to classified military data. Arms control frameworks do not cover the data inputs of AI targeting systems. The consequence is that the most consequential dataset produced by the most AI-intensive conflict in history will be used to train future military AI systems under conditions of no governance, no transparency, and no accountability, by the states with the resources and access to exploit it, against the states that do not.
*The views presented in this article are the authors’ own and do not necessarily reflect the views of The Diplomatic Insight.
Aleena Saif Ullah
Aleena Saif Ullahis an MPhil Scholar in International Relations at the University of Punjab, Lahore. She can be reached at[email protected]
